HENT-SRT (#2026)

HENT-SRT: Hierarchical Efficient Neural Transducer with Self-Distillation for Joint Speech Recognition and Translation

Paper: https://arxiv.org/abs/2506.02157

egs/multi_conv_zh_es_ta/ST/README.md

@@ -0,0 +1,53 @@
+# HENT-SRT
+
+This repository contains a **speech-to-text translation (ST)** recipe accompanying our IWSLT 2025 paper:
+
+**HENT-SRT: Hierarchical Efficient Neural Transducer with Self-Distillation for Joint Speech Recognition and Translation**  
+Paper: <https://arxiv.org/abs/2506.02157>
+
+## Datasets
+
+The recipe combines three conversational, 3-way parallel ST corpora:
+
+- **Tunisian–English (IWSLT’22 TA)**  
+  Lhotse recipe: <https://github.com/lhotse-speech/lhotse/blob/master/lhotse/recipes/iwslt22_ta.py>
+
+- **Fisher Spanish**  
+  Reference: <https://aclanthology.org/2013.iwslt-papers.14>
+
+- **HKUST (Mandarin Telephone Speech)**  
+  Reference: <https://arxiv.org/abs/2404.11619>
+
+> **Data access:** Fisher and HKUST require an institutional LDC subscription.  
+> **Recipe status:** Lhotse recipes for Fisher Spanish and HKUST are in progress and will be finalized soon.
+
+
+
+## Zipformer Multi-joiner ST
+This model is similar to https://www.isca-archive.org/interspeech_2023/wang23oa_interspeech.pdf, but 
+our system uses zipformer encoder with a pruned transducer and stateless decoder
+
+| Dataset    | Decoding method      | test WER | test BLEU | comment                                         |
+| ---------- | -------------------- | -------- | --------- | ----------------------------------------------- |
+| iwslt\_ta  | modified beam search | 41.6      | 16.3       | --epoch 20, --avg 13, beam(20),  |
+| hkust      | modified beam search | 23.8      | 10.4       | --epoch 20, --avg 13, beam(20),  |
+| fisher\_sp | modified beam search | 18.0      | 31.0       | --epoch 20, --avg 13, beam(20),  |
+
+
+## Hent-SRT offline
+
+| Dataset    | Decoding method      | test WER | test BLEU | comment                                         |
+| ---------- | -------------------- | -------- | --------- | ----------------------------------------------- |
+| iwslt\_ta  | modified beam search | 41.4      | 20.6       | --epoch 20, --avg 13, beam(20),  BP 1 |
+| hkust      | modified beam search | 22.8      | 14.7       | --epoch 20, --avg 13, beam(20),  BP 1 |
+| fisher\_sp | modified beam search | 17.8      | 33.7       | --epoch 20, --avg 13, beam(20),  BP 1 |
+
+## Hent-SRT streaming
+
+| Dataset    | Decoding method      | test WER | test BLEU | comment                                         |
+| ---------- | -------------------- | -------- | --------- | ----------------------------------------------- |
+| iwslt\_ta  | greedy search | 46.2      | 17.3       | --epoch 20, --avg 13, BP 2, chunk-size 64, left-context-frames 128, max-sym-per-frame 20  |
+| hkust      | greedy search | 27.3      | 11.2       | --epoch 20, --avg 13, BP 2, chunk-size 64, left-context-frames 128, max-sym-per-frame 20|
+| fisher\_sp | greedy search | 22.7      | 30.8     | --epoch 20, --avg 13, BP 2, chunk-size 64, left-context-frames 128, max-sym-per-frame 20 |
+
+See [RESULTS](/egs/multi_conv_zh_es_ta/ST/RESULTS.md) for details.

egs/multi_conv_zh_es_ta/ST/RESULTS.md

@@ -0,0 +1,404 @@
+## Zipformer Multi-joiner ST
+
+
+### For offline model training:  
+
+You can find a pretrained model, training logs, decoding logs, and decoding results at: https://huggingface.co/AmirHussein/HENT-SRT/tree/main/zipformer_multijoiner_st
+
+
+| Dataset    | Decoding method      | test WER | test BLEU | comment                                         |
+| ---------- | -------------------- | -------- | --------- | ----------------------------------------------- |
+| iwslt\_ta  | modified beam search | 41.6      | 16.3       | --epoch 25, --avg 13, beam(20),  |
+| hkust      | modified beam search | 23.8      | 10.4       | --epoch 25, --avg 13, beam(20),  |
+| fisher\_sp | modified beam search | 18.0      | 31.0       | --epoch 25, --avg 13, beam(20),  |
+
+The training command:
+
+```
+export CUDA_VISIBLE_DEVICES="0,1,2,3"
+./zipformer_multijoiner_st/train.py \
+  --base-lr 0.045 \
+  --world-size 4 \
+  --num-epochs 30 \
+  --start-epoch 1 \
+  --use-fp16 1 \
+  --exp-dir zipformer_multijoiner_st/exp-multi-joiner-pbe4k\
+  --causal 0 \
+  --num-encoder-layers 2,2,2,2,2,2,2,2 \
+  --feedforward-dim 512,768,1024,1024,1024,1024,1024,768 \
+  --encoder-dim 192,256,384,512,384,384,384,256 \
+  --encoder-unmasked-dim 192,192,256,256,256,256,256,192 \
+  --downsampling-factor 1,2,4,8,8,4,4,2\
+  --cnn-module-kernel 31,31,15,15,15,15,31,31 \
+  --num-heads 4,4,4,8,8,8,4,4 \
+  --bpe-st-model data/lang_st_bpe_4000/bpe.model \
+  --max-duration 400 \
+  --prune-range 10 \
+  --warm-step 10000 \
+  --lr-epochs 6 \
+  --use-hat False
+  ```
+
+Decodeing command:
+```
+ ./zipformer_multijoiner_st/decode.py \
+        --exp-dir ./zipformer_multijoiner_st/exp-multi-joiner-pbe4k \
+        --epoch 25 \
+        --avg 13 \
+        --beam-size 20 \
+        --max-duration 600 \
+        --decoding-method modified_beam_search \
+        --bpe-st-model data/lang_st_bpe_4000/bpe.model \
+        --bpe-model data/lang_bpe_5000/bpe.model \
+        --num-encoder-layers 2,2,2,2,2,2,2,2 \
+        --feedforward-dim  512,768,1024,1024,1024,1024,1024,768 \
+        --encoder-dim  192,256,384,512,384,384,384,256 \
+        --encoder-unmasked-dim 192,192,256,256,256,256,256,192 \
+        --downsampling-factor 1,2,4,8,8,4,4,2 \
+        --cnn-module-kernel 31,31,15,15,15,15,31,31 \
+        --num-heads 4,4,4,8,8,8,4,4 \
+        --use-averaged-model True
+```
+
+### For streaming  model training:  
+
+| Dataset    | Decoding method      | test WER | test BLEU | comment                                         |
+| ---------- | -------------------- | -------- | --------- | ----------------------------------------------- |
+| iwslt\_ta  | greedy search |   44.1    |    6.0    | --epoch 25, --avg 13  |
+| hkust      | greedy search |    27.4  |    3.7   | --epoch 25, --avg 13  |
+| fisher\_sp | greedy search |  19.9     |  16.3      | --epoch 25, --avg 13  |
+
+The training command:
+
+```
+export CUDA_VISIBLE_DEVICES="0,1,2,3"
+./zipformer_multijoiner_st/train.py \
+  --base-lr 0.045 \
+  --world-size 4 \
+  --num-epochs 30 \
+  --start-epoch 1 \
+  --use-fp16 1 \
+  --exp-dir zipformer_multijoiner_st/exp-multi-joiner-pbe4k_causal\
+  --causal 1 \
+  --num-encoder-layers 2,2,2,2,2,2,2,2 \
+  --feedforward-dim 512,768,1024,1024,1024,1024,1024,768 \
+  --encoder-dim 192,256,384,512,384,384,384,256 \
+  --encoder-unmasked-dim 192,192,256,256,256,256,256,192 \
+  --downsampling-factor 1,2,4,8,8,4,4,2\
+  --cnn-module-kernel 31,31,15,15,15,15,31,31 \
+  --num-heads 4,4,4,8,8,8,4,4 \
+  --bpe-st-model data/lang_st_bpe_4000/bpe.model \
+  --max-duration 400 \
+  --prune-range 10 \
+  --warm-step 10000 \
+  --lr-epochs 6 \
+  --use-hat False
+  ```
+
+Decodeing command:
+```
+ ./zipformer_multijoiner_st/decode.py \
+        --exp-dir ./zipformer_multijoiner_st/exp-multi-joiner-pbe4k \
+        --causal 1  \
+        --epoch 25 \
+        --avg 13 \
+        --beam-size 20 \
+        --max-duration 600 \
+        --decoding-method modified_beam_search \
+        --bpe-st-model data/lang_st_bpe_4000/bpe.model \
+        --bpe-model data/lang_bpe_5000/bpe.model \
+        --num-encoder-layers 2,2,2,2,2,2,2,2 \
+        --feedforward-dim  512,768,1024,1024,1024,1024,1024,768 \
+        --encoder-dim  192,256,384,512,384,384,384,256 \
+        --encoder-unmasked-dim 192,192,256,256,256,256,256,192 \
+        --downsampling-factor 1,2,4,8,8,4,4,2 \
+        --cnn-module-kernel 31,31,15,15,15,15,31,31 \
+        --num-heads 4,4,4,8,8,8,4,4 \
+        --use-averaged-model True \
+        --decoding-method greedy_search \
+        --chunk-size 64 \
+        --left-context-frames 128 \
+        --use-hat False \
+        --max-sym-per-frame 20 
+```
+
+
+## Hent-SRT offline
+
+You can find a pretrained model, training logs, decoding logs, and decoding results at: https://huggingface.co/AmirHussein/HENT-SRT/tree/main/hent_srt
+
+| Dataset    | Decoding method      | test WER | test BLEU | comment                                         |
+| ---------- | -------------------- | -------- | --------- | ----------------------------------------------- |
+| iwslt\_ta  | modified beam search | 41.4      | 20.6       | --epoch 20, --avg 13, beam(20),  BP 1 |
+| hkust      | modified beam search | 22.8      | 14.7       | --epoch 20, --avg 13, beam(20),  BP 1 |
+| fisher\_sp | modified beam search | 17.8      | 33.7       | --epoch 20, --avg 13, beam(20),  BP 1 |
+
+
+### First pretrain the offline CR-CTC ASR
+
+```
+export CUDA_VISIBLE_DEVICES="0,1,2,3"
+./hent_srt/train.py \
+    --base-lr 0.055 \
+    --world-size 4 \
+    --num-epochs 30 \
+    --start-epoch 1 \
+    --use-fp16 1 \
+    --exp-dir hent_srt/exp-asr\
+    --causal 0 \
+    --num-encoder-layers 2,2,2,2,2 \
+    --feedforward-dim 512,768,1024,1024,1024 \
+    --encoder-dim 192,256,384,512,384 \
+    --encoder-unmasked-dim 192,192,256,256,256 \
+    --downsampling-factor 1,2,4,8,4 \
+    --cnn-module-kernel 31,31,15,15,15 \
+    --num-heads 4,4,4,8,8 \
+    --st-num-encoder-layers 2,2,2,2,2 \
+    --st-feedforward-dim 512,512,256,256,256 \
+    --st-encoder-dim 512,384,256,256,256 \
+    --st-encoder-unmasked-dim 256,256,256,256,192 \
+    --st-downsampling-factor 4,4,4,4,4 \
+    --st-cnn-module-kernel 15,31,31,15,15 \
+    --st-num-heads 4,4,8,8,8 \
+    --bpe-st-model data/lang_st_bpe_4000/bpe.model \
+    --bpe-model data/lang_bpe_5000/bpe.model \
+    --manifest-dir data/fbank \
+    --max-duration 350 \
+    --prune-range 10 \
+    --warm-step 8000 \
+    --ctc-loss-scale 0.2 \
+    --enable-spec-aug 0 \
+    --cr-loss-scale 0.2 \
+    --time-mask-ratio 2.5 \
+    --use-asr-cr-ctc 1 \
+    --use-ctc 1 \
+    --lr-epochs 6 \
+    --use-hat False \
+    --use-st-joiner False 
+```
+
+### Train ST with a Pretrained ASR Initialization
+
+```
+export CUDA_VISIBLE_DEVICES="0,1,2,3"
+./hent_srt/train.py \
+  --base-lr 0.045 \
+  --world-size 4 \
+  --num-epochs 25 \
+  --start-epoch 1 \
+  --use-fp16 1 \
+  --exp-dir hent_srt/exp-st \
+  --model-init-ckpt hent_srt/exp-asr/best-valid-loss.pt \
+  --causal 0 \
+  --num-encoder-layers 2,2,2,2,2 \
+  --feedforward-dim 512,768,1024,1024,1024 \
+  --encoder-dim 192,256,384,512,384 \
+  --encoder-unmasked-dim 192,192,256,256,256 \
+  --downsampling-factor 1,2,4,8,4 \
+  --cnn-module-kernel 31,31,15,15,15 \
+  --num-heads 4,4,4,8,8 \
+  --st-num-encoder-layers 2,2,2,2,2 \
+  --st-feedforward-dim 512,512,256,256,256 \
+  --st-encoder-dim 384,512,256,256,256 \
+  --st-encoder-unmasked-dim 256,256,256,256,192 \
+  --st-downsampling-factor 1,2,4,4,4  \
+  --st-cnn-module-kernel 15,31,31,15,15 \
+  --st-num-heads 8,8,8,8,8 \
+  --output-downsampling-factor 2 \
+  --st-output-downsampling-factor 1 \
+  --bpe-st-model data/lang_st_bpe_4000/bpe.model \
+  --bpe-model data/lang_bpe_5000/bpe.model \
+  --manifest-dir data/fbank \
+  --max-duration 200 \
+  --prune-range 5 \
+  --st-prune-range 10 \
+  --warm-step 10000 \
+  --ctc-loss-scale 0.1 \
+  --st-ctc-loss-scale 0.1 \
+  --enable-spec-aug 0 \
+  --cr-loss-scale 0.05 \
+  --st-cr-loss-scale 0.05 \
+  --time-mask-ratio 2.5 \
+  --use-asr-cr-ctc 1 \
+  --use-ctc 1 \
+  --use-st-cr-ctc 1 \
+  --use-st-ctc 1 \
+  --lr-epochs 6 \
+  --use-hat False \
+  --use-st-joiner True
+```
+
+### Decode offline Hent-SRT
+
+```
+./hent_srt/decode.py \
+    --epoch 20 --avg 13 --use-averaged-model True \
+    --beam-size 20 \
+    --causal 0 \
+    --exp-dir hent_srt/exp-st \
+    --bpe-model data/lang_bpe_5000/bpe.model \
+    --bpe-st-model data/lang_st_bpe_4000/bpe.model \
+    --output-downsampling-factor 2 \
+    --st-output-downsampling-factor 1 \
+    --max-duration 800 \
+    --num-encoder-layers 2,2,2,2,2 \
+    --feedforward-dim 512,768,1024,1024,1024 \
+    --encoder-dim 192,256,384,512,384 \
+    --encoder-unmasked-dim 192,192,256,256,256 \
+    --downsampling-factor 1,2,4,8,4 \
+    --cnn-module-kernel 31,31,15,15,15 \
+    --num-heads 4,4,4,8,8 \
+    --st-num-encoder-layers 2,2,2,2,2 \
+    --st-feedforward-dim 512,512,256,256,256 \
+    --st-encoder-dim 384,512,256,256,256 \
+    --st-encoder-unmasked-dim 256,256,256,256,192 \
+    --st-downsampling-factor 1,2,4,4,4  \
+    --st-cnn-module-kernel 15,31,31,15,15 \
+    --st-num-heads 8,8,8,8,8 \
+    --decoding-method modified_beam_search \
+    --use-st-joiner True \
+    --use-hat-decode False \
+    --use-ctc 1 \
+    --use-st-ctc 1 \
+    --st-blank-penalty 1
+```
+
+
+## Hent-SRT streaming
+
+| Dataset    | Decoding method      | test WER | test BLEU | comment                                         |
+| ---------- | -------------------- | -------- | --------- | ----------------------------------------------- |
+| iwslt\_ta  | greedy search | 46.2      | 17.3       | --epoch 20, --avg 13, BP 2, chunk-size 64, left-context-frames 128, max-sym-per-frame 20  |
+| hkust      | greedy search | 27.3      | 11.2       | --epoch 20, --avg 13, BP 2, chunk-size 64, left-context-frames 128, max-sym-per-frame 20|
+| fisher\_sp | greedy search | 22.7      | 30.8     | --epoch 20, --avg 13, BP 2, chunk-size 64, left-context-frames 128, max-sym-per-frame 20 |
+
+### First pretrain the streaming CR-CTC ASR
+
+# CR-CTC ASR streaming
+```
+export CUDA_VISIBLE_DEVICES="0,1,2,3"
+./hent_srt/train.py \
+    --base-lr 0.055 \
+    --world-size 4 \
+    --num-epochs 30 \
+    --start-epoch 1 \
+    --use-fp16 1 \
+    --exp-dir hent_srt/exp-asr_causal\
+    --causal 1 \
+    --num-encoder-layers 2,2,2,2,2 \
+    --feedforward-dim 512,768,1024,1024,1024 \
+    --encoder-dim 192,256,384,512,384 \
+    --encoder-unmasked-dim 192,192,256,256,256 \
+    --downsampling-factor 1,2,4,8,4 \
+    --cnn-module-kernel 31,31,15,15,15 \
+    --num-heads 4,4,4,8,8 \
+    --st-num-encoder-layers 2,2,2,2,2 \
+    --st-feedforward-dim 512,512,256,256,256 \
+    --st-encoder-dim 512,384,256,256,256 \
+    --st-encoder-unmasked-dim 256,256,256,256,192 \
+    --st-downsampling-factor 4,4,4,4,4 \
+    --st-cnn-module-kernel 15,31,31,15,15 \
+    --st-num-heads 4,4,8,8,8 \
+    --bpe-st-model data/lang_st_bpe_4000/bpe.model \
+    --bpe-model data/lang_bpe_5000/bpe.model \
+    --manifest-dir data/fbank \
+    --max-duration 250 \
+    --prune-range 10 \
+    --warm-step 8000 \
+    --ctc-loss-scale 0.2 \
+    --enable-spec-aug 0 \
+    --cr-loss-scale 0.2 \
+    --time-mask-ratio 2.5 \
+    --use-asr-cr-ctc 1 \
+    --use-ctc 1 \
+    --lr-epochs 6 \
+    --use-hat False \
+    --use-st-joiner False 
+```
+
+
+### Train streaming ST with a Pretrained ASR Initialization
+```
+export CUDA_VISIBLE_DEVICES="0,1,2,3"
+./hent_srt/train.py \
+  --base-lr 0.045 \
+  --world-size 4 \
+  --num-epochs 25 \
+  --start-epoch 1 \
+  --use-fp16 1 \
+  --exp-dir hent_srt/exp-st_causal \
+  --model-init-ckpt hent_srt/exp-asr_causal/best-valid-loss.pt \
+  --causal 1 \
+  --num-encoder-layers 2,2,2,2,2 \
+  --feedforward-dim 512,768,1024,1024,1024 \
+  --encoder-dim 192,256,384,512,384 \
+  --encoder-unmasked-dim 192,192,256,256,256 \
+  --downsampling-factor 1,2,4,8,4 \
+  --cnn-module-kernel 31,31,15,15,15 \
+  --num-heads 4,4,4,8,8 \
+  --st-num-encoder-layers 2,2,2,2,2 \
+  --st-feedforward-dim 512,512,256,256,256 \
+  --st-encoder-dim 384,512,256,256,256 \
+  --st-encoder-unmasked-dim 256,256,256,256,192 \
+  --st-downsampling-factor 1,2,4,4,4  \
+  --st-cnn-module-kernel 15,31,31,15,15 \
+  --st-num-heads 8,8,8,8,8 \
+  --output-downsampling-factor 2 \
+  --st-output-downsampling-factor 1 \
+  --bpe-st-model data/lang_st_bpe_4000/bpe.model \
+  --bpe-model data/lang_bpe_5000/bpe.model \
+  --manifest-dir data/fbank \
+  --max-duration 200 \
+  --prune-range 5 \
+  --st-prune-range 10 \
+  --warm-step 10000 \
+  --ctc-loss-scale 0.1 \
+  --st-ctc-loss-scale 0.1 \
+  --enable-spec-aug 0 \
+  --cr-loss-scale 0.05 \
+  --st-cr-loss-scale 0.05 \
+  --time-mask-ratio 2.5 \
+  --use-asr-cr-ctc 1 \
+  --use-ctc 1 \
+  --use-st-cr-ctc 1 \
+  --use-st-ctc 1 \
+  --lr-epochs 6 \
+  --use-hat False \
+  --use-st-joiner True 
+  ```
+
+  ### Decode streaming Hent-SRT
+```
+./hent_srt/decode.py \
+        --epoch 20 --avg 13 --use-averaged-model True \
+        --causal 1 \
+        --exp-dir hent_srt/exp-st_causal \
+        --bpe-model data/lang_bpe_5000/bpe.model \
+        --bpe-st-model data/lang_st_bpe_4000/bpe.model \
+        --output-downsampling-factor 2 \
+        --st-output-downsampling-factor 1 \
+        --max-duration 800 \
+        --num-encoder-layers 2,2,2,2,2 \
+        --feedforward-dim 512,768,1024,1024,1024 \
+        --encoder-dim 192,256,384,512,384 \
+        --encoder-unmasked-dim 192,192,256,256,256 \
+        --downsampling-factor 1,2,4,8,4 \
+        --cnn-module-kernel 31,31,15,15,15 \
+        --num-heads 4,4,4,8,8 \
+        --st-num-encoder-layers 2,2,2,2,2 \
+        --st-feedforward-dim 512,512,256,256,256 \
+        --st-encoder-dim 384,512,256,256,256 \
+        --st-encoder-unmasked-dim 256,256,256,256,192 \
+        --st-downsampling-factor 1,2,4,4,4  \
+        --st-cnn-module-kernel 15,31,31,15,15 \
+        --st-num-heads 8,8,8,8,8 \
+        --decoding-method greedy_search \
+        --use-st-joiner True \
+        --use-hat-decode False \
+        --use-ctc 1 \
+        --use-st-ctc 1 \
+        --st-blank-penalty 2 \
+        --chunk-size 64 \
+        --left-context-frames 128 \
+        --use-hat False --max-sym-per-frame 20 
+```

egs/multi_conv_zh_es_ta/ST/hent_srt/asr_datamodule.py

@@ -0,0 +1 @@
+../zipformer_multijoiner_st/asr_datamodule.py

egs/multi_conv_zh_es_ta/ST/hent_srt/decoder.py

@@ -0,0 +1,261 @@
+# Copyright      2025  Johns Hopkins University (author: Amir Hussein)
+#
+# See ../../../../LICENSE for clarification regarding multiple authors
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+from torch.nn.utils.rnn import pad_sequence
+
+from scaling import Balancer
+
+
+# Copyright    2021-2024  Xiaomi Corp.        (authors: Fangjun Kuang,
+#                                                       Zengrui Jin,
+#                                                       Yifan Yang,)
+#
+# See ../../../../LICENSE for clarification regarding multiple authors
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from typing import Optional, Tuple
+
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+from scaling import Balancer
+
+
+class LSTMDecoder(nn.Module):
+    """LSTM decoder."""
+
+    def __init__(
+        self,
+        vocab_size: int,
+        blank_id: int,
+        decoder_dim: int,
+        num_layers: int,
+        hidden_dim: int,
+        embedding_dropout: float = 0.0,
+        rnn_dropout: float = 0.0,
+    ):
+        """
+        Args:
+          vocab_size:
+            Number of tokens of the modeling unit including blank.
+          blank_id:
+            The ID of the blank symbol.
+          decoder_dim:
+            Dimension of the input embedding.
+          num_layers:
+            Number of LSTM layers.
+          hidden_dim:
+            Hidden dimension of LSTM layers.
+          embedding_dropout:
+            Dropout rate for the embedding layer.
+          rnn_dropout:
+            Dropout for LSTM layers.
+        """
+        super().__init__()
+
+        self.embedding = nn.Embedding(
+            num_embeddings=vocab_size,
+            embedding_dim=decoder_dim,
+        )
+        # the balancers are to avoid any drift in the magnitude of the
+        # embeddings, which would interact badly with parameter averaging.
+        self.balancer = Balancer(
+            decoder_dim,
+            channel_dim=-1,
+            min_positive=0.0,
+            max_positive=1.0,
+            min_abs=0.5,
+            max_abs=1.0,
+            prob=0.05,
+        )
+
+        self.blank_id = blank_id
+
+        self.vocab_size = vocab_size
+
+        self.embedding_dropout = nn.Dropout(embedding_dropout)
+
+        self.rnn = nn.LSTM(
+            input_size=decoder_dim,
+            hidden_size=hidden_dim,
+            num_layers=num_layers,
+            batch_first=True,
+            dropout=rnn_dropout,
+        )
+  
+        self.balancer2 = Balancer(
+            decoder_dim,
+            channel_dim=-1,
+            min_positive=0.0,
+            max_positive=1.0,
+            min_abs=0.5,
+            max_abs=1.0,
+            prob=0.05,
+        )
+
+    def forward(
+        self,
+        y: torch.Tensor,
+        states: Optional[Tuple[torch.Tensor, torch.Tensor]] = None,
+        need_pad: bool = False
+    ) -> torch.Tensor:
+        """
+        Args:
+          y:
+            A 2-D tensor of shape (N, U).
+        Returns:
+          Return a tensor of shape (N, U, decoder_dim).
+        """
+        y = y.to(torch.int64)
+        # this stuff about clamp() is a temporary fix for a mismatch
+        # at utterance start, we use negative ids in beam_search.py
+        embedding_out = self.embedding(y.clamp(min=0)) * (y >= 0).unsqueeze(-1)
+
+        embedding_out = self.embedding_dropout(embedding_out)
+
+        embedding_out = self.balancer(embedding_out)
+        if need_pad is True:
+                embedding_out = pad_sequence(embedding_out, batch_first=True, padding_value=0)
+
+        rnn_out, (h, c) = self.rnn(embedding_out, states)
+
+        rnn_out = F.relu(rnn_out)
+        rnn_out = self.balancer2(rnn_out)
+
+        return rnn_out, (h, c)
+
+
+class Decoder(nn.Module):
+    """This class modifies the stateless decoder from the following paper:
+
+        RNN-transducer with stateless prediction network
+        https://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=9054419
+
+    It removes the recurrent connection from the decoder, i.e., the prediction
+    network. Different from the above paper, it adds an extra Conv1d
+    right after the embedding layer.
+
+    TODO: Implement https://arxiv.org/pdf/2109.07513.pdf
+    """
+
+    def __init__(
+        self,
+        vocab_size: int,
+        decoder_dim: int,
+        blank_id: int,
+        context_size: int,
+    ):
+        """
+        Args:
+          vocab_size:
+            Number of tokens of the modeling unit including blank.
+          decoder_dim:
+            Dimension of the input embedding, and of the decoder output.
+          blank_id:
+            The ID of the blank symbol.
+          context_size:
+            Number of previous words to use to predict the next word.
+            1 means bigram; 2 means trigram. n means (n+1)-gram.
+        """
+        super().__init__()
+
+        self.embedding = nn.Embedding(
+            num_embeddings=vocab_size,
+            embedding_dim=decoder_dim,
+        )
+        # the balancers are to avoid any drift in the magnitude of the
+        # embeddings, which would interact badly with parameter averaging.
+        self.balancer = Balancer(
+            decoder_dim,
+            channel_dim=-1,
+            min_positive=0.0,
+            max_positive=1.0,
+            min_abs=0.5,
+            max_abs=1.0,
+            prob=0.05,
+        )
+
+        self.blank_id = blank_id
+
+        assert context_size >= 1, context_size
+        self.context_size = context_size
+        self.vocab_size = vocab_size
+
+        if context_size > 1:
+            self.conv = nn.Conv1d(
+                in_channels=decoder_dim,
+                out_channels=decoder_dim,
+                kernel_size=context_size,
+                padding=0,
+                groups=decoder_dim // 4,  # group size == 4
+                bias=False,
+            )
+            self.balancer2 = Balancer(
+                decoder_dim,
+                channel_dim=-1,
+                min_positive=0.0,
+                max_positive=1.0,
+                min_abs=0.5,
+                max_abs=1.0,
+                prob=0.05,
+            )
+
+    def forward(self, y: torch.Tensor, need_pad: bool = True) -> torch.Tensor:
+        """
+        Args:
+          y:
+            A 2-D tensor of shape (N, U).
+          need_pad:
+            True to left pad the input. Should be True during training.
+            False to not pad the input. Should be False during inference.
+        Returns:
+          Return a tensor of shape (N, U, decoder_dim).
+        """
+        y = y.to(torch.int64)
+        # this stuff about clamp() is a temporary fix for a mismatch
+        # at utterance start, we use negative ids in beam_search.py
+        embedding_out = self.embedding(y.clamp(min=0)) * (y >= 0).unsqueeze(-1)
+
+        embedding_out = self.balancer(embedding_out)
+
+        if self.context_size > 1:
+            embedding_out = embedding_out.permute(0, 2, 1)
+            if need_pad is True:
+                embedding_out = F.pad(embedding_out, pad=(self.context_size - 1, 0))
+            else:
+                # During inference time, there is no need to do extra padding
+                # as we only need one output
+                assert embedding_out.size(-1) == self.context_size
+            embedding_out = self.conv(embedding_out)
+            embedding_out = embedding_out.permute(0, 2, 1)
+            embedding_out = F.relu(embedding_out)
+            embedding_out = self.balancer2(embedding_out)
+
+        return embedding_out

egs/multi_conv_zh_es_ta/ST/hent_srt/export.py

@@ -0,0 +1,640 @@
+#!/usr/bin/env python3
+#
+# Copyright      2025  Johns Hopkins University (author: Amir Hussein)
+# Copyright 2021-2023 Xiaomi Corporation (Author: Fangjun Kuang,
+#                                                 Zengwei Yao,
+#                                                 Wei Kang)
+#
+# See ../../../../LICENSE for clarification regarding multiple authors
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+# This script converts several saved checkpoints
+# to a single one using model averaging.
+"""
+
+Usage:
+
+Note: This is a example for librispeech dataset, if you are using different
+dataset, you should change the argument values according to your dataset.
+
+(1) Export to torchscript model using torch.jit.script()
+
+- For non-streaming model:
+
+./hent_srt/export.py \
+    --exp-dir ./hent_srt/exp-st \
+    --causal 0 \
+    --use-averaged-model 1 \
+    --tokens data/lang_bpe_5000/tokens.txt \
+    --st-tokens data/lang_st_bpe_4000/tokens.txt \
+    --num-encoder-layers 2,2,2,2,2 \
+    --feedforward-dim 512,768,1024,1024,1024 \
+    --encoder-dim 192,256,384,512,384 \
+    --encoder-unmasked-dim 192,192,256,256,256 \
+    --downsampling-factor 1,2,4,8,4 \
+    --cnn-module-kernel 31,31,15,15,15 \
+    --num-heads 4,4,4,8,8 \
+    --st-num-encoder-layers 2,2,2,2,2 \
+    --st-feedforward-dim 512,512,256,256,256 \
+    --st-encoder-dim 384,512,256,256,256 \
+    --st-encoder-unmasked-dim 256,256,256,256,192 \
+    --st-downsampling-factor 1,2,4,4,4  \
+    --st-cnn-module-kernel 15,31,31,15,15 \
+    --st-num-heads 8,8,8,8,8 \
+    --epoch 25 \
+    --avg 13 \
+    --jit 1 \
+    --output-downsampling-factor 2 \
+    --st-output-downsampling-factor 1 \
+    --use-st-joiner True \
+    --use-hat-decode False \
+    --use-ctc 1 \
+    --use-st-ctc 1
+
+It will generate a file `jit_script.pt` in the given `exp_dir`. You can later
+load it by `torch.jit.load("jit_script.pt")`.
+
+Check ./jit_pretrained.py for its usage.
+
+Check https://github.com/k2-fsa/sherpa
+for how to use the exported models outside of icefall.
+
+- For streaming model:
+
+./zipformer/export.py \
+  --exp-dir ./zipformer/exp \
+  --causal 1 \
+  --chunk-size 32 \
+  --left-context-frames 128 \
+  --tokens data/lang_bpe_500/tokens.txt \
+  --epoch 30 \
+  --avg 9 \
+  --jit 1
+
+It will generate a file `jit_script_chunk_16_left_128.pt` in the given `exp_dir`.
+You can later load it by `torch.jit.load("jit_script_chunk_16_left_128.pt")`.
+
+Check ./jit_pretrained_streaming.py for its usage.
+
+Check https://github.com/k2-fsa/sherpa
+for how to use the exported models outside of icefall.
+
+(2) Export `model.state_dict()`
+
+- For non-streaming model:
+
+./hent_srt/export.py \
+    --exp-dir ./hent_srt/exp-st \
+    --causal 0 \
+    --use-averaged-model 1 \
+    --tokens data/lang_bpe_5000/tokens.txt \
+    --st-tokens data/lang_st_bpe_4000/tokens.txt \
+    --num-encoder-layers 2,2,2,2,2 \
+    --feedforward-dim 512,768,1024,1024,1024 \
+    --encoder-dim 192,256,384,512,384 \
+    --encoder-unmasked-dim 192,192,256,256,256 \
+    --downsampling-factor 1,2,4,8,4 \
+    --cnn-module-kernel 31,31,15,15,15 \
+    --num-heads 4,4,4,8,8 \
+    --st-num-encoder-layers 2,2,2,2,2 \
+    --st-feedforward-dim 512,512,256,256,256 \
+    --st-encoder-dim 384,512,256,256,256 \
+    --st-encoder-unmasked-dim 256,256,256,256,192 \
+    --st-downsampling-factor 1,2,4,4,4  \
+    --st-cnn-module-kernel 15,31,31,15,15 \
+    --st-num-heads 8,8,8,8,8 \
+    --epoch 20 \
+    --avg 13 \
+    --jit 0 \
+    --output-downsampling-factor 2 \
+    --st-output-downsampling-factor 1 \
+    --use-st-joiner True \
+    --use-hat False \
+    --use-ctc 1 \
+    --use-st-ctc 1
+
+- For streaming model:
+
+./hent_srt/export.py \
+  --exp-dir ./hent_srt/exp-st_causal \
+    --causal 1 \
+    --use-averaged-model 1 \
+    --tokens data/lang_bpe_5000/tokens.txt \
+    --st-tokens data/lang_st_bpe_4000/tokens.txt \
+    --num-encoder-layers 2,2,2,2,2 \
+    --feedforward-dim 512,768,1024,1024,1024 \
+    --encoder-dim 192,256,384,512,384 \
+    --encoder-unmasked-dim 192,192,256,256,256 \
+    --downsampling-factor 1,2,4,8,4 \
+    --cnn-module-kernel 31,31,15,15,15 \
+    --num-heads 4,4,4,8,8 \
+    --st-num-encoder-layers 2,2,2,2,2 \
+    --st-feedforward-dim 512,512,256,256,256 \
+    --st-encoder-dim 384,512,256,256,256 \
+    --st-encoder-unmasked-dim 256,256,256,256,192 \
+    --st-downsampling-factor 1,2,4,4,4  \
+    --st-cnn-module-kernel 15,31,31,15,15 \
+    --st-num-heads 8,8,8,8,8 \
+    --epoch 20 \
+    --avg 13 \
+    --jit 0 \
+    --output-downsampling-factor 2 \
+    --st-output-downsampling-factor 1 \
+    --use-st-joiner True \
+    --use-hat False \
+    --use-ctc 1 \
+    --use-st-ctc 1
+
+It will generate a file `pretrained.pt` in the given `exp_dir`. You can later
+load it by `icefall.checkpoint.load_checkpoint()`.
+
+- For non-streaming model:
+
+To use the generated file with `zipformer/decode.py`,
+you can do:
+
+    cd /path/to/exp_dir
+    ln -s pretrained.pt epoch-9999.pt
+
+    cd /path/to/egs/multi_conv_zh_es_ta/ST
+    
+    ./hent_srt/decode.py \
+        --epoch 9999 --avg 1 --use-averaged-model 0 \
+        --beam-size 20 \
+        --causal 0 \
+        --exp-dir hent_srt/exp-st \
+        --bpe-model data/lang_bpe_5000/bpe.model \
+        --bpe-st-model data/lang_st_bpe_4000/bpe.model \
+        --output-downsampling-factor 2 \
+        --st-output-downsampling-factor 1 \
+        --max-duration 800 \
+        --num-encoder-layers 2,2,2,2,2 \
+        --feedforward-dim 512,768,1024,1024,1024 \
+        --encoder-dim 192,256,384,512,384 \
+        --encoder-unmasked-dim 192,192,256,256,256 \
+        --downsampling-factor 1,2,4,8,4 \
+        --cnn-module-kernel 31,31,15,15,15 \
+        --num-heads 4,4,4,8,8 \
+        --st-num-encoder-layers 2,2,2,2,2 \
+        --st-feedforward-dim 512,512,256,256,256 \
+        --st-encoder-dim 384,512,256,256,256 \
+        --st-encoder-unmasked-dim 256,256,256,256,192 \
+        --st-downsampling-factor 1,2,4,4,4  \
+        --st-cnn-module-kernel 15,31,31,15,15 \
+        --st-num-heads 8,8,8,8,8 \
+        --decoding-method modified_beam_search \
+        --use-st-joiner True \
+        --use-hat-decode False \
+        --use-ctc 1 \
+        --use-st-ctc 1 \
+        --st-blank-penalty 1
+
+- For streaming model:
+
+To use the generated file with `zipformer/decode.py` and `zipformer/streaming_decode.py`, you can do:
+
+    cd /path/to/exp_dir
+    ln -s pretrained.pt epoch-9999.pt
+
+    cd /path/to/egs/multi_conv_zh_es_ta/ST
+        
+    ./hent_srt/decode.py \
+        --epoch 9999 --avg 1 --use-averaged-model 0 \
+        --causal 1 \
+        --exp-dir hent_srt/exp-st_causal \
+        --bpe-model data/lang_bpe_5000/bpe.model \
+        --bpe-st-model data/lang_st_bpe_4000/bpe.model \
+        --output-downsampling-factor 2 \
+        --st-output-downsampling-factor 1 \
+        --max-duration 800 \
+        --num-encoder-layers 2,2,2,2,2 \
+        --feedforward-dim 512,768,1024,1024,1024 \
+        --encoder-dim 192,256,384,512,384 \
+        --encoder-unmasked-dim 192,192,256,256,256 \
+        --downsampling-factor 1,2,4,8,4 \
+        --cnn-module-kernel 31,31,15,15,15 \
+        --num-heads 4,4,4,8,8 \
+        --st-num-encoder-layers 2,2,2,2,2 \
+        --st-feedforward-dim 512,512,256,256,256 \
+        --st-encoder-dim 384,512,256,256,256 \
+        --st-encoder-unmasked-dim 256,256,256,256,192 \
+        --st-downsampling-factor 1,2,4,4,4  \
+        --st-cnn-module-kernel 15,31,31,15,15 \
+        --st-num-heads 8,8,8,8,8 \
+        --decoding-method greedy_search \
+        --use-st-joiner True \
+        --use-hat-decode False \
+        --use-ctc 1 \
+        --use-st-ctc 1 \
+        --st-blank-penalty 2 \
+        --chunk-size 64 \
+        --left-context-frames 128 \
+        --use-hat False --max-sym-per-frame 20 
+
+Note: If you don't want to train a model from scratch, we have
+provided one for you. You can get it at
+
+- non-streaming model:
+https://huggingface.co/Zengwei/icefall-asr-librispeech-zipformer-2023-05-15
+
+- streaming model:
+https://huggingface.co/Zengwei/icefall-asr-librispeech-streaming-zipformer-2023-05-17
+
+with the following commands:
+
+    sudo apt-get install git-lfs
+    git lfs install
+    git clone https://huggingface.co/Zengwei/icefall-asr-librispeech-zipformer-2023-05-15
+    git clone https://huggingface.co/Zengwei/icefall-asr-librispeech-streaming-zipformer-2023-05-17
+    # You will find the pre-trained models in exp dir
+"""
+
+import argparse
+import logging
+from pathlib import Path
+from typing import List, Tuple
+
+import k2
+import torch
+from scaling_converter import convert_scaled_to_non_scaled
+from torch import Tensor, nn
+from train import add_model_arguments, get_model, get_params
+
+from icefall.checkpoint import (
+    average_checkpoints,
+    average_checkpoints_with_averaged_model,
+    find_checkpoints,
+    load_checkpoint,
+)
+from icefall.utils import make_pad_mask, num_tokens, str2bool
+
+
+def get_parser():
+    parser = argparse.ArgumentParser(
+        formatter_class=argparse.ArgumentDefaultsHelpFormatter
+    )
+
+    parser.add_argument(
+        "--epoch",
+        type=int,
+        default=30,
+        help="""It specifies the checkpoint to use for decoding.
+        Note: Epoch counts from 1.
+        You can specify --avg to use more checkpoints for model averaging.""",
+    )
+
+    parser.add_argument(
+        "--iter",
+        type=int,
+        default=0,
+        help="""If positive, --epoch is ignored and it
+        will use the checkpoint exp_dir/checkpoint-iter.pt.
+        You can specify --avg to use more checkpoints for model averaging.
+        """,
+    )
+
+    parser.add_argument(
+        "--avg",
+        type=int,
+        default=9,
+        help="Number of checkpoints to average. Automatically select "
+        "consecutive checkpoints before the checkpoint specified by "
+        "'--epoch' and '--iter'",
+    )
+
+    parser.add_argument(
+        "--use-averaged-model",
+        type=str2bool,
+        default=True,
+        help="Whether to load averaged model. Currently it only supports "
+        "using --epoch. If True, it would decode with the averaged model "
+        "over the epoch range from `epoch-avg` (excluded) to `epoch`."
+        "Actually only the models with epoch number of `epoch-avg` and "
+        "`epoch` are loaded for averaging. ",
+    )
+
+    parser.add_argument(
+        "--exp-dir",
+        type=str,
+        default="zipformer/exp",
+        help="""It specifies the directory where all training related
+        files, e.g., checkpoints, log, etc, are saved
+        """,
+    )
+
+    parser.add_argument(
+        "--tokens",
+        type=str,
+        default="data/lang_bpe_5000/tokens.txt",
+        help="Path to the tokens.txt",
+    )
+    parser.add_argument(
+        "--st-tokens",
+        type=str,
+        default="data/lang_st_bpe_4000/tokens.txt",
+        help="Path to the tokens.txt",
+    )
+
+    parser.add_argument(
+        "--jit",
+        type=str2bool,
+        default=False,
+        help="""True to save a model after applying torch.jit.script.
+        It will generate a file named jit_script.pt.
+        Check ./jit_pretrained.py for how to use it.
+        """,
+    )
+
+    parser.add_argument(
+        "--context-size",
+        type=int,
+        default=2,
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
+    )
+    parser.add_argument(
+        "--st-context-size",
+        type=int,
+        default=2,
+        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+    )
+
+    add_model_arguments(parser)
+
+    return parser
+
+
+class EncoderModel(nn.Module):
+    """A wrapper for encoder and encoder_embed"""
+
+    def __init__(self, model: nn.Module) -> None:
+        super().__init__()
+        self.model = model
+
+    def forward(
+        self, features: Tensor, feature_lengths: Tensor
+    ) -> Tuple[Tensor, Tensor]:
+        """
+        Args:
+            features: (N, T, C)
+            feature_lengths: (N,)
+        """
+        encoder_out, encoder_out_lens, st_encoder_out, st_encoder_out_lens = model.forward_encoder(feature, feature_lengths)
+        encoder_out = encoder_out.permute(1, 0, 2)  # (T, N, C) ->(N, T, C)
+        st_encoder_out = st_encoder_out.permute(1, 0, 2)
+        return encoder_out, encoder_out_lens, st_encoder_out, st_encoder_out_lens
+
+
+class StreamingEncoderModel(nn.Module):
+    """A wrapper for encoder and encoder_embed"""
+
+    def __init__(self, encoder: nn.Module, encoder_embed: nn.Module) -> None:
+        super().__init__()
+        assert len(encoder.chunk_size) == 1, encoder.chunk_size
+        assert len(encoder.left_context_frames) == 1, encoder.left_context_frames
+        self.chunk_size = encoder.chunk_size[0]
+        self.left_context_len = encoder.left_context_frames[0]
+
+        # The encoder_embed subsample features (T - 7) // 2
+        # The ConvNeXt module needs (7 - 1) // 2 = 3 frames of right padding after subsampling
+        self.pad_length = 7 + 2 * 3
+
+        self.encoder = encoder
+        self.encoder_embed = encoder_embed
+
+    def forward(
+        self, features: Tensor, feature_lengths: Tensor, states: List[Tensor]
+    ) -> Tuple[Tensor, Tensor, List[Tensor]]:
+        """Streaming forward for encoder_embed and encoder.
+
+        Args:
+            features: (N, T, C)
+            feature_lengths: (N,)
+            states: a list of Tensors
+
+        Returns encoder outputs, output lengths, and updated states.
+        """
+        chunk_size = self.chunk_size
+        left_context_len = self.left_context_len
+
+        cached_embed_left_pad = states[-2]
+        x, x_lens, new_cached_embed_left_pad = self.encoder_embed.streaming_forward(
+            x=features,
+            x_lens=feature_lengths,
+            cached_left_pad=cached_embed_left_pad,
+        )
+        assert x.size(1) == chunk_size, (x.size(1), chunk_size)
+
+        src_key_padding_mask = make_pad_mask(x_lens)
+
+        # processed_mask is used to mask out initial states
+        processed_mask = torch.arange(left_context_len, device=x.device).expand(
+            x.size(0), left_context_len
+        )
+        processed_lens = states[-1]  # (batch,)
+        # (batch, left_context_size)
+        processed_mask = (processed_lens.unsqueeze(1) <= processed_mask).flip(1)
+        # Update processed lengths
+        new_processed_lens = processed_lens + x_lens
+
+        # (batch, left_context_size + chunk_size)
+        src_key_padding_mask = torch.cat([processed_mask, src_key_padding_mask], dim=1)
+
+        x = x.permute(1, 0, 2)  # (N, T, C) -> (T, N, C)
+        encoder_states = states[:-2]
+
+        (
+            encoder_out,
+            encoder_out_lens,
+            new_encoder_states,
+        ) = self.encoder.streaming_forward(
+            x=x,
+            x_lens=x_lens,
+            states=encoder_states,
+            src_key_padding_mask=src_key_padding_mask,
+        )
+        encoder_out = encoder_out.permute(1, 0, 2)  # (T, N, C) ->(N, T, C)
+
+        new_states = new_encoder_states + [
+            new_cached_embed_left_pad,
+            new_processed_lens,
+        ]
+        return encoder_out, encoder_out_lens, new_states
+
+    @torch.jit.export
+    def get_init_states(
+        self,
+        batch_size: int = 1,
+        device: torch.device = torch.device("cpu"),
+    ) -> List[torch.Tensor]:
+        """
+        Returns a list of cached tensors of all encoder layers. For layer-i, states[i*6:(i+1)*6]
+        is (cached_key, cached_nonlin_attn, cached_val1, cached_val2, cached_conv1, cached_conv2).
+        states[-2] is the cached left padding for ConvNeXt module,
+        of shape (batch_size, num_channels, left_pad, num_freqs)
+        states[-1] is processed_lens of shape (batch,), which records the number
+        of processed frames (at 50hz frame rate, after encoder_embed) for each sample in batch.
+        """
+        states = self.encoder.get_init_states(batch_size, device)
+
+        embed_states = self.encoder_embed.get_init_states(batch_size, device)
+        states.append(embed_states)
+
+        processed_lens = torch.zeros(batch_size, dtype=torch.int32, device=device)
+        states.append(processed_lens)
+
+        return states
+
+
+@torch.no_grad()
+def main():
+    args = get_parser().parse_args()
+    args.exp_dir = Path(args.exp_dir)
+
+    params = get_params()
+    params.update(vars(args))
+
+    device = torch.device("cpu")
+    # if torch.cuda.is_available():
+    #     device = torch.device("cuda", 0)
+
+    logging.info(f"device: {device}")
+
+    token_table = k2.SymbolTable.from_file(params.tokens)
+    st_token_table = k2.SymbolTable.from_file(params.st_tokens)
+    params.blank_id = token_table["<blk>"]
+    # params.unk_id = sp.piece_to_id("<unk>")
+    # params.st_unk_id = sp_st.piece_to_id("<unk>")
+    params.blank_st_id = st_token_table["<blk>"]
+    params.vocab_size = num_tokens(token_table) + 1
+    params.vocab_st_size = num_tokens(st_token_table) + 1
+
+    logging.info(params)
+
+    logging.info("About to create model")
+    model = get_model(params)
+
+    if not params.use_averaged_model:
+        if params.iter > 0:
+            filenames = find_checkpoints(params.exp_dir, iteration=-params.iter)[
+                : params.avg
+            ]
+            if len(filenames) == 0:
+                raise ValueError(
+                    f"No checkpoints found for"
+                    f" --iter {params.iter}, --avg {params.avg}"
+                )
+            elif len(filenames) < params.avg:
+                raise ValueError(
+                    f"Not enough checkpoints ({len(filenames)}) found for"
+                    f" --iter {params.iter}, --avg {params.avg}"
+                )
+            logging.info(f"averaging {filenames}")
+            model.load_state_dict(average_checkpoints(filenames, device=device))
+        elif params.avg == 1:
+            load_checkpoint(f"{params.exp_dir}/epoch-{params.epoch}.pt", model)
+        else:
+            start = params.epoch - params.avg + 1
+            filenames = []
+            for i in range(start, params.epoch + 1):
+                if i >= 1:
+                    filenames.append(f"{params.exp_dir}/epoch-{i}.pt")
+            logging.info(f"averaging {filenames}")
+            model.load_state_dict(average_checkpoints(filenames, device=device))
+    else:
+        if params.iter > 0:
+            filenames = find_checkpoints(params.exp_dir, iteration=-params.iter)[
+                : params.avg + 1
+            ]
+            if len(filenames) == 0:
+                raise ValueError(
+                    f"No checkpoints found for"
+                    f" --iter {params.iter}, --avg {params.avg}"
+                )
+            elif len(filenames) < params.avg + 1:
+                raise ValueError(
+                    f"Not enough checkpoints ({len(filenames)}) found for"
+                    f" --iter {params.iter}, --avg {params.avg}"
+                )
+            filename_start = filenames[-1]
+            filename_end = filenames[0]
+            logging.info(
+                "Calculating the averaged model over iteration checkpoints"
+                f" from {filename_start} (excluded) to {filename_end}"
+            )
+            model.load_state_dict(
+                average_checkpoints_with_averaged_model(
+                    filename_start=filename_start,
+                    filename_end=filename_end,
+                    device=device,
+                )
+            )
+        elif params.avg == 1:
+            load_checkpoint(f"{params.exp_dir}/epoch-{params.epoch}.pt", model)
+        else:
+            assert params.avg > 0, params.avg
+            start = params.epoch - params.avg
+            assert start >= 1, start
+            filename_start = f"{params.exp_dir}/epoch-{start}.pt"
+            filename_end = f"{params.exp_dir}/epoch-{params.epoch}.pt"
+            logging.info(
+                f"Calculating the averaged model over epoch range from "
+                f"{start} (excluded) to {params.epoch}"
+            )
+            model.load_state_dict(
+                average_checkpoints_with_averaged_model(
+                    filename_start=filename_start,
+                    filename_end=filename_end,
+                    device=device,
+                    
+                ), strict=False
+            )
+
+    model.eval()
+
+    if params.jit is True:
+        convert_scaled_to_non_scaled(model, inplace=True)
+        # We won't use the forward() method of the model in C++, so just ignore
+        # it here.
+        # Otherwise, one of its arguments is a ragged tensor and is not
+        # torch scriptabe.
+        model.__class__.forward = torch.jit.ignore(model.__class__.forward)
+
+        # Wrap encoder and encoder_embed as a module
+        if params.causal:
+            model.encoder = StreamingEncoderModel(model.encoder, model.encoder_embed)
+            chunk_size = model.encoder.chunk_size
+            left_context_len = model.encoder.left_context_len
+            filename = f"jit_script_chunk_{chunk_size}_left_{left_context_len}.pt"
+        else:
+            model.encoder = EncoderModel(model)
+            filename = "jit_script.pt"
+
+        logging.info("Using torch.jit.script")
+        model = torch.jit.script(model)
+        model.save(str(params.exp_dir / filename))
+        logging.info(f"Saved to {filename}")
+    else:
+        logging.info("Not using torchscript. Export model.state_dict()")
+        # Save it using a format so that it can be loaded
+        # by :func:`load_checkpoint`
+        filename = params.exp_dir / "pretrained.pt"
+        torch.save({"model": model.state_dict()}, str(filename))
+        logging.info(f"Saved to {filename}")
+
+
+if __name__ == "__main__":
+    formatter = "%(asctime)s %(levelname)s [%(filename)s:%(lineno)d] %(message)s"
+
+    logging.basicConfig(format=formatter, level=logging.INFO)
+    main()

egs/multi_conv_zh_es_ta/ST/hent_srt/joiner.py

@@ -0,0 +1 @@
+../zipformer_multijoiner_st/joiner.py

egs/multi_conv_zh_es_ta/ST/hent_srt/load_pretrained_model.py

@@ -0,0 +1,117 @@
+# Copyright      2025  Johns Hopkins University (author: Amir Hussein)
+
+import logging
+from typing import Any, Dict, Union
+
+import torch
+import torch.nn
+import torch.optim
+
+
+def filter_state_dict(
+    dst_state: Dict[str, Union[float, torch.Tensor]],
+    src_state: Dict[str, Union[float, torch.Tensor]],
+):
+    """Filter name, size mismatch instances between dicts.
+
+    Args:
+        dst_state: reference state dict for filtering
+        src_state: target state dict for filtering
+
+    """
+    match_state = {}
+    for key, value in src_state.items():
+        if key in dst_state and (dst_state[key].size() == src_state[key].size()):
+            match_state[key] = value
+        else:
+            if key not in dst_state:
+                logging.warning(
+                    f"Filter out {key} from pretrained dict"
+                    + " because of name not found in target dict"
+                )
+            else:
+                logging.warning(
+                    f"Filter out {key} from pretrained dict"
+                    + " because of size mismatch"
+                    + f"({dst_state[key].size()}-{src_state[key].size()})"
+                )
+    return match_state
+
+
+def load_pretrained_model(
+    init_param: str,
+    model: torch.nn.Module,
+    ignore_init_mismatch: bool,
+    map_location: str = "cpu",
+):
+    """Load a model state and set it to the model.
+
+    Args:
+        init_param: <file_path>:<src_key>:<dst_key>:<exclude_Keys>
+
+    Examples:
+        >>> load_pretrained_model("somewhere/model.pth", model)
+        >>> load_pretrained_model("somewhere/model.pth:decoder:decoder", model)
+        >>> load_pretrained_model("somewhere/model.pth:decoder:decoder:", model)
+        >>> load_pretrained_model(
+        ...     "somewhere/model.pth:decoder:decoder:decoder.embed", model
+        ... )
+        >>> load_pretrained_model("somewhere/decoder.pth::decoder", model)
+    """
+    sps = init_param.split(":", 4)
+    if len(sps) == 4:
+        path, src_key, dst_key, excludes = sps
+    elif len(sps) == 3:
+        path, src_key, dst_key = sps
+        excludes = None
+    elif len(sps) == 2:
+        path, src_key = sps
+        dst_key, excludes = None, None
+    else:
+        (path,) = sps
+        src_key, dst_key, excludes = None, None, None
+    if src_key == "":
+        src_key = None
+    if dst_key == "":
+        dst_key = None
+
+    if dst_key is None:
+        obj = model
+    else:
+
+        def get_attr(obj: Any, key: str):
+            """Get an nested attribute.
+
+            >>> class A(torch.nn.Module):
+            ...     def __init__(self):
+            ...         super().__init__()
+            ...         self.linear = torch.nn.Linear(10, 10)
+            >>> a = A()
+            >>> assert A.linear.weight is get_attr(A, 'linear.weight')
+
+            """
+            
+            if key.strip() == "":
+                return obj
+            for k in key.split("."):
+                obj = getattr(obj, k)
+            return obj
+
+        obj = get_attr(model, dst_key)
+    
+    src_state = torch.load(path, map_location=map_location)
+    
+    if excludes is not None:
+        for e in excludes.split(","):
+            src_state = {k: v for k, v in src_state.items() if not k.startswith(e)}
+    if src_key is not None:
+        src_state['model'] = {
+            k[len(src_key) + 1 :]: v
+            for k, v in src_state['model'].items()
+            if k.startswith(src_key)
+        }
+    dst_state = obj.state_dict()
+    if ignore_init_mismatch:
+        src_state = filter_state_dict(dst_state, src_state['model'])
+    dst_state.update(src_state)
+    obj.load_state_dict(dst_state)

egs/multi_conv_zh_es_ta/ST/hent_srt/model.py

@@ -0,0 +1,811 @@
+# Copyright    2025  Johns Hopkins University (author: Amir Hussein)
+# Copyright    2021-2023  Xiaomi Corp.        (authors: Fangjun Kuang,
+#                                                       Wei Kang,
+#                                                       Zengwei Yao)
+#
+# See ../../../../LICENSE for clarification regarding multiple authors
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from typing import Optional, Tuple, Union
+
+import k2
+import torch
+from torch import Tensor
+from lhotse.dataset import SpecAugment
+import torch.nn as nn
+from encoder_interface import EncoderInterface
+from scaling import ScaledLinear
+from icefall.utils import add_sos, make_pad_mask, time_warp
+
+
+class HENT_SRT(nn.Module):
+    def __init__(
+        self,
+        encoder_embed: nn.Module,
+        encoder: EncoderInterface,
+        decoder: Optional[nn.Module] = None,
+        joiner: Optional[nn.Module] = None,
+        st_joiner: Optional[nn.Module] = None,
+        st_decoder: Optional[nn.Module] = None,
+        st_encoder: Optional[nn.Module] = None,
+        encoder_dim: int = 384,
+        st_encoder_dim: int = 384,
+        decoder_dim: int = 512,
+        vocab_size: int = 500,
+        st_vocab_size: int = 500,
+        use_transducer: bool = True,
+        use_ctc: bool = False,
+        use_st_ctc: bool = False,
+        use_hat: bool = False,
+        use_lstm_pred:bool=False,
+    ):
+        """A multitask Transducer ASR-ST model with seperate joiners and predictors but shared acoustic encoder.
+
+        - Connectionist temporal classification: labelling unsegmented sequence data with recurrent neural networks (http://imagine.enpc.fr/~obozinsg/teaching/mva_gm/papers/ctc.pdf)
+        - Sequence Transduction with Recurrent Neural Networks (https://arxiv.org/pdf/1211.3711.pdf)
+        - Pruned RNN-T for fast, memory-efficient ASR training (https://arxiv.org/pdf/2206.13236.pdf)
+
+        Args:
+          encoder_embed:
+            It is a Convolutional 2D subsampling module. It converts
+            an input of shape (N, T, idim) to an output of of shape
+            (N, T', odim), where T' = (T-3)//2-2 = (T-7)//2.
+          encoder:
+            It is the transcription network in the paper. Its accepts
+            two inputs: `x` of (N, T, encoder_dim) and `x_lens` of shape (N,).
+            It returns two tensors: `logits` of shape (N, T, encoder_dim) and
+            `logit_lens` of shape (N,).
+          decoder:
+            It is the prediction network in the paper. Its input shape
+            is (N, U) and its output shape is (N, U, decoder_dim).
+            It should contain one attribute: `blank_id`.
+            It is used when use_transducer is True.
+          joiner:
+            It has two inputs with shapes: (N, T, encoder_dim) and (N, U, decoder_dim).
+            Its output shape is (N, T, U, vocab_size). Note that its output contains
+            unnormalized probs, i.e., not processed by log-softmax.
+            It is used when use_transducer is True.
+          use_transducer:
+            Whether use transducer head. Default: True.
+          use_ctc:
+            Whether use CTC head. Default: False.
+        """
+        super().__init__()
+
+        assert (
+            use_transducer or use_ctc
+        ), f"At least one of them should be True, but got use_transducer={use_transducer}, use_ctc={use_ctc}"
+
+        assert isinstance(encoder, EncoderInterface), type(encoder)
+
+        self.encoder_embed = encoder_embed
+        self.encoder = encoder
+        self.use_hat = use_hat
+        self.use_lstm_pred = use_lstm_pred
+        self.use_transducer = use_transducer
+        if use_transducer:
+            # Modules for Transducer head
+            assert decoder is not None
+            assert hasattr(decoder, "blank_id")
+            assert joiner is not None
+
+            self.decoder = decoder
+            self.joiner = joiner
+            
+            self.st_joiner = st_joiner
+            self.st_decoder = st_decoder
+            self.st_encoder = st_encoder
+
+            self.simple_am_proj = ScaledLinear(
+                encoder_dim, vocab_size, initial_scale=0.25
+            )
+            self.simple_lm_proj = ScaledLinear(
+                decoder_dim, vocab_size, initial_scale=0.25
+            )
+
+            self.simple_st_am_proj = ScaledLinear(
+                st_encoder_dim, st_vocab_size, initial_scale=0.25
+            )
+            self.simple_st_lm_proj = ScaledLinear(
+                decoder_dim, st_vocab_size, initial_scale=0.25
+            )
+        else:
+            assert decoder is None
+            assert joiner is None
+
+        self.use_ctc = use_ctc
+        self.use_st_ctc = use_st_ctc
+        if self.use_ctc:
+            # Modules for CTC head
+            self.ctc_output = nn.Sequential(
+                nn.Dropout(p=0.1),
+                nn.Linear(encoder_dim, vocab_size),
+                nn.LogSoftmax(dim=-1),
+            )
+        if self.use_st_ctc:
+            # Modules for CTC head
+            self.st_ctc_output = nn.Sequential(
+                nn.Dropout(p=0.1),
+                nn.Linear(st_encoder_dim, st_vocab_size),
+                nn.LogSoftmax(dim=-1),
+            )
+
+    def forward_encoder(
+        self, x: torch.Tensor, x_lens: torch.Tensor
+    ) -> Tuple[torch.Tensor, torch.Tensor]:
+        """Compute encoder outputs.
+        Args:
+          x:
+            A 3-D tensor of shape (N, T, C).
+          x_lens:
+            A 1-D tensor of shape (N,). It contains the number of frames in `x`
+            before padding.
+
+        Returns:
+          encoder_out:
+            Encoder output, of shape (N, T, C).
+          encoder_out_lens:
+            Encoder output lengths, of shape (N,).
+        """
+        # logging.info(f"Memory allocated at entry: {torch.cuda.memory_allocated() // 1000000}M")
+        x, x_lens = self.encoder_embed(x, x_lens)
+        # logging.info(f"Memory allocated after encoder_embed: {torch.cuda.memory_allocated() // 1000000}M")
+
+        src_key_padding_mask = make_pad_mask(x_lens)
+        x = x.permute(1, 0, 2)  # (N, T, C) -> (T, N, C)
+
+        encoder_out, encoder_out_lens, st_input = self.encoder(x, x_lens, src_key_padding_mask)
+        if self.st_encoder is not None:
+          st_src_key_padding_mask = make_pad_mask(encoder_out_lens)
+          
+          st_encoder_out, st_encoder_out_lens = self.st_encoder(
+                st_input, x_lens, src_key_padding_mask
+            )
+          st_encoder_out = st_encoder_out.permute(1, 0, 2)  # (T, N, C) ->(N, T, C)
+        else:
+           st_encoder_out_lens = None
+           st_encoder_out = None
+        encoder_out = encoder_out.permute(1, 0, 2)  # (T, N, C) ->(N, T, C)
+        assert torch.all(encoder_out_lens > 0), (x_lens, encoder_out_lens)
+        return encoder_out, encoder_out_lens, st_encoder_out, st_encoder_out_lens
+
+    def forward_st_ctc(
+        self,
+        st_encoder_out: torch.Tensor,
+        st_encoder_out_lens: torch.Tensor,
+        targets: torch.Tensor,
+        target_lengths: torch.Tensor,
+    ) -> torch.Tensor:
+        """Compute CTC loss.
+        Args:
+          encoder_out:
+            Encoder output, of shape (N, T, C).
+          encoder_out_lens:
+            Encoder output lengths, of shape (N,).
+          targets:
+            Target Tensor of shape (sum(target_lengths)). The targets are assumed
+            to be un-padded and concatenated within 1 dimension.
+        """
+        # Compute CTC log-prob
+        ctc_output = self.st_ctc_output(st_encoder_out)  # (N, T, C)
+
+        ctc_loss = torch.nn.functional.ctc_loss(
+            log_probs=ctc_output.permute(1, 0, 2),  # (T, N, C)
+            targets=targets,
+            input_lengths=st_encoder_out_lens,
+            target_lengths=target_lengths,
+            reduction="sum",
+        )
+        return ctc_loss
+
+    def forward_ctc(
+        self,
+        encoder_out: torch.Tensor,
+        encoder_out_lens: torch.Tensor,
+        targets: torch.Tensor,
+        target_lengths: torch.Tensor,
+    ) -> torch.Tensor:
+        """Compute CTC loss.
+        Args:
+          encoder_out:
+            Encoder output, of shape (N, T, C).
+          encoder_out_lens:
+            Encoder output lengths, of shape (N,).
+          targets:
+            Target Tensor of shape (sum(target_lengths)). The targets are assumed
+            to be un-padded and concatenated within 1 dimension.
+        """
+        # Compute CTC log-prob
+        ctc_output = self.ctc_output(encoder_out)  # (N, T, C)
+
+        ctc_loss = torch.nn.functional.ctc_loss(
+            log_probs=ctc_output.permute(1, 0, 2),  # (T, N, C)
+            targets=targets,
+            input_lengths=encoder_out_lens,
+            target_lengths=target_lengths,
+            reduction="sum",
+        )
+        return ctc_loss
+
+    def forward_cr_ctc(
+        self,
+        encoder_out: torch.Tensor,
+        encoder_out_lens: torch.Tensor,
+        targets: torch.Tensor,
+        target_lengths: torch.Tensor,
+    ) -> Tuple[torch.Tensor, torch.Tensor]:
+        """Compute CTC loss with consistency regularization loss.
+        Args:
+          encoder_out:
+            Encoder output, of shape (2 * N, T, C).
+          encoder_out_lens:
+            Encoder output lengths, of shape (2 * N,).
+          targets:
+            Target Tensor of shape (2 * sum(target_lengths)). The targets are assumed
+            to be un-padded and concatenated within 1 dimension.
+        """
+        # Compute CTC loss
+        ctc_output = self.ctc_output(encoder_out)  # (2 * N, T, C)
+        ctc_loss = torch.nn.functional.ctc_loss(
+            log_probs=ctc_output.permute(1, 0, 2),  # (T, 2 * N, C)
+            targets=targets.cpu(),
+            input_lengths=encoder_out_lens.cpu(),
+            target_lengths=target_lengths.cpu(),
+            reduction="none",
+        )
+        ctc_loss_is_finite = torch.isfinite(ctc_loss)
+        ctc_loss = ctc_loss[ctc_loss_is_finite]
+        ctc_loss = ctc_loss.sum()
+
+        # Compute consistency regularization loss
+        exchanged_targets = ctc_output.detach().chunk(2, dim=0)
+        exchanged_targets = torch.cat(
+            [exchanged_targets[1], exchanged_targets[0]], dim=0
+        )  # exchange: [x1, x2] -> [x2, x1]
+        cr_loss = nn.functional.kl_div(
+            input=ctc_output,
+            target=exchanged_targets,
+            reduction="none",
+            log_target=True,
+        )  # (2 * N, T, C)
+        length_mask = make_pad_mask(encoder_out_lens).unsqueeze(-1)
+        cr_loss = cr_loss.masked_fill(length_mask, 0.0).sum()
+
+        return ctc_loss, cr_loss
+    
+    def forward_st_cr_ctc(
+        self,
+        st_encoder_out: torch.Tensor,
+        st_encoder_out_lens: torch.Tensor,
+        st_targets: torch.Tensor,
+        st_target_lengths: torch.Tensor,
+        # encoder_out: torch.Tensor,
+        # encoder_out_lens: torch.Tensor,
+        # targets: torch.Tensor,
+        # target_lengths: torch.Tensor,
+    ) -> Tuple[torch.Tensor, torch.Tensor]:
+        """Compute CTC loss with consistency regularization loss.
+        Args:
+          encoder_out:
+            Encoder output, of shape (2 * N, T, C).
+          encoder_out_lens:
+            Encoder output lengths, of shape (2 * N,).
+          targets:
+            Target Tensor of shape (2 * sum(target_lengths)). The targets are assumed
+            to be un-padded and concatenated within 1 dimension.
+        """
+        # Compute CTC loss
+        st_ctc_output = self.st_ctc_output(st_encoder_out)  # (2 * N, T, C)
+        st_ctc_loss = torch.nn.functional.ctc_loss(
+            log_probs=st_ctc_output.permute(1, 0, 2),  # (T, 2 * N, C)
+            targets=st_targets.cpu(),
+            input_lengths=st_encoder_out_lens.cpu(),
+            target_lengths=st_target_lengths.cpu(),
+            reduction="none",
+        )
+        st_ctc_loss_is_finite = torch.isfinite(st_ctc_loss)
+        st_ctc_loss = st_ctc_loss[st_ctc_loss_is_finite]
+        st_ctc_loss = st_ctc_loss.sum()
+        # ctc_output = self.ctc_output(encoder_out)  # (2 * N, T, C)
+        # ctc_loss = torch.nn.functional.ctc_loss(
+        #     log_probs=ctc_output.permute(1, 0, 2),  # (T, 2 * N, C)
+        #     targets=targets.cpu(),
+        #     input_lengths=encoder_out_lens.cpu(),
+        #     target_lengths=target_lengths.cpu(),
+        #     reduction="sum",
+        # )
+        
+        # if not torch.isfinite(st_ctc_loss):
+        #             breakpoint()
+
+        # Compute consistency regularization loss
+        exchanged_targets = st_ctc_output.detach().chunk(2, dim=0)
+        exchanged_targets = torch.cat(
+            [exchanged_targets[1], exchanged_targets[0]], dim=0
+        )  # exchange: [x1, x2] -> [x2, x1]
+        cr_loss = nn.functional.kl_div(
+            input=st_ctc_output,
+            target=exchanged_targets,
+            reduction="none",
+            log_target=True,
+        )  # (2 * N, T, C)
+        length_mask = make_pad_mask(st_encoder_out_lens).unsqueeze(-1)
+        cr_loss = cr_loss.masked_fill(length_mask, 0.0).sum()
+
+        return st_ctc_loss, cr_loss
+
+    def forward_st_transducer(
+        self,
+        encoder_out: torch.Tensor,
+        encoder_out_lens: torch.Tensor,
+        st_encoder_out: torch.Tensor,
+        st_encoder_out_lens: torch.Tensor,
+        y: k2.RaggedTensor,
+        y_lens: torch.Tensor,
+        st_y: k2.RaggedTensor,
+        st_y_lens: torch.Tensor,
+        prune_range: int = 5,
+        st_prune_range: int = 10,
+        am_scale: float = 0.0,
+        lm_scale: float = 0.0,
+    ) -> Union[Tuple[Tensor, Tensor], Tuple[Tensor, Tensor, Tensor]]:
+        """Compute Transducer loss.
+        Args:
+          encoder_out:
+            Encoder output, of shape (N, T, C).
+          encoder_out_lens:
+            Encoder output lengths, of shape (N,).
+          y:
+            A ragged tensor with 2 axes [utt][label]. It contains labels of each
+            utterance.
+          prune_range:
+            The prune range for rnnt loss, it means how many symbols(context)
+            we are considering for each frame to compute the loss.
+          am_scale:
+            The scale to smooth the loss with am (output of encoder network)
+            part
+          lm_scale:
+            The scale to smooth the loss with lm (output of predictor network)
+            part
+        """
+        # Now for the decoder, i.e., the prediction network
+        
+
+        blank_id = self.decoder.blank_id
+        st_blank_id = self.st_decoder.blank_id
+        sos_y = add_sos(y, sos_id=blank_id)
+        st_sos_y = add_sos(st_y, sos_id=st_blank_id)
+
+        # sos_y_padded: [B, S + 1], start with SOS.
+        sos_y_padded = sos_y.pad(mode="constant", padding_value=blank_id)
+        st_sos_y_padded = st_sos_y.pad(mode="constant", padding_value=st_blank_id)
+
+        # decoder_out: [B, S + 1, decoder_dim]
+        decoder_out = self.decoder(sos_y_padded)
+        if self.use_lstm_pred:
+          st_decoder_out, _ = self.st_decoder(st_sos_y_padded)
+        else:
+          st_decoder_out = self.st_decoder(st_sos_y_padded)
+
+        # Note: y does not start with SOS
+        # y_padded : [B, S]
+        y_padded = y.pad(mode="constant", padding_value=0)
+
+        y_padded = y_padded.to(torch.int64)
+        boundary = torch.zeros(
+            (encoder_out.size(0), 4),
+            dtype=torch.int64,
+            device=encoder_out.device,
+        )
+        boundary[:, 2] = y_lens
+        boundary[:, 3] = encoder_out_lens
+
+        st_y_padded = st_y.pad(mode="constant", padding_value=0)
+
+        st_y_padded = st_y_padded.to(torch.int64)
+        st_boundary = torch.zeros(
+            (encoder_out.size(0), 4),
+            dtype=torch.int64,
+            device=encoder_out.device,
+        )
+        st_boundary[:, 2] = st_y_lens
+        st_boundary[:, 3] = st_encoder_out_lens
+
+        lm = self.simple_lm_proj(decoder_out)
+        am = self.simple_am_proj(encoder_out)
+        
+        st_lm = self.simple_st_lm_proj(st_decoder_out)
+        st_am = self.simple_st_am_proj(st_encoder_out)
+
+        # if self.training and random.random() < 0.25:
+        #    lm = penalize_abs_values_gt(lm, 100.0, 1.0e-04)
+        # if self.training and random.random() < 0.25:
+        #    am = penalize_abs_values_gt(am, 30.0, 1.0e-04)
+
+        with torch.cuda.amp.autocast(enabled=False):
+          simple_loss, (px_grad, py_grad) = k2.rnnt_loss_smoothed(
+              lm=lm.float(),
+              am=am.float(),
+              symbols=y_padded,
+              termination_symbol=blank_id,
+              lm_only_scale=lm_scale,
+              am_only_scale=am_scale,
+              boundary=boundary,
+              reduction="sum",
+              return_grad=True,
+          )
+          st_simple_loss, (st_px_grad, st_py_grad) = k2.rnnt_loss_smoothed(
+              lm=st_lm.float(),
+              am=st_am.float(),
+              symbols=st_y_padded,
+              termination_symbol=st_blank_id,
+              lm_only_scale=lm_scale,
+              am_only_scale=am_scale,
+              boundary=st_boundary,
+              reduction="sum",
+              return_grad=True,
+          )
+          
+
+        # am_pruned : [B, T, prune_range, encoder_dim]
+        # lm_pruned : [B, T, prune_range, decoder_dim]
+       
+        # ranges : [B, T, prune_range]
+        ranges = k2.get_rnnt_prune_ranges(
+            px_grad=px_grad,
+            py_grad=py_grad,
+            boundary=boundary,
+            s_range=prune_range,
+        )
+        am_pruned, lm_pruned = k2.do_rnnt_pruning(
+            am=self.joiner.encoder_proj(encoder_out),
+            lm=self.joiner.decoder_proj(decoder_out),
+            ranges=ranges,
+        )
+          # project_input=False since we applied the decoder's input projections
+          # prior to do_rnnt_pruning (this is an optimization for speed).
+        logits = self.joiner(am_pruned, lm_pruned, project_input=False)
+        with torch.cuda.amp.autocast(enabled=False):
+          pruned_loss = k2.rnnt_loss_pruned(
+              logits=logits.float(),
+              symbols=y_padded,
+              ranges=ranges,
+              termination_symbol=blank_id,
+              boundary=boundary,
+              reduction="sum",
+              use_hat_loss=self.use_hat,
+          )
+         # logits : [B, T, prune_range, vocab_size]
+        
+        st_ranges = k2.get_rnnt_prune_ranges(
+          px_grad=st_px_grad,
+          py_grad=st_py_grad,
+          boundary=st_boundary,
+          s_range=st_prune_range,
+        )
+        st_am_pruned, st_lm_pruned = k2.do_rnnt_pruning(
+            am=self.st_joiner.encoder_proj(st_encoder_out),
+            lm=self.st_joiner.decoder_proj(st_decoder_out),
+            ranges=st_ranges,
+        )
+
+        st_logits = self.st_joiner(st_am_pruned, st_lm_pruned, project_input=False)
+        # Compute HAT loss for st
+        with torch.cuda.amp.autocast(enabled=False):
+          pruned_st_loss = k2.rnnt_loss_pruned(
+              logits=st_logits.float(),
+              symbols=st_y.pad(mode="constant", padding_value=blank_id).to(torch.int64),
+              ranges=st_ranges,
+              termination_symbol=st_blank_id,
+              boundary=st_boundary,
+              reduction="sum",
+              use_hat_loss=self.use_hat,
+          )
+        
+        return simple_loss, st_simple_loss, pruned_loss, pruned_st_loss
+
+
+    def forward_transducer(
+        self,
+        encoder_out: torch.Tensor,
+        encoder_out_lens: torch.Tensor,
+        y: k2.RaggedTensor,
+        y_lens: torch.Tensor,
+        prune_range: int = 5,
+        am_scale: float = 0.0,
+        lm_scale: float = 0.0,
+    ) -> Union[Tuple[Tensor, Tensor], Tuple[Tensor, Tensor, Tensor]]:
+        """Compute Transducer loss.
+        Args:
+          encoder_out:
+            Encoder output, of shape (N, T, C).
+          encoder_out_lens:
+            Encoder output lengths, of shape (N,).
+          y:
+            A ragged tensor with 2 axes [utt][label]. It contains labels of each
+            utterance.
+          prune_range:
+            The prune range for rnnt loss, it means how many symbols(context)
+            we are considering for each frame to compute the loss.
+          am_scale:
+            The scale to smooth the loss with am (output of encoder network)
+            part
+          lm_scale:
+            The scale to smooth the loss with lm (output of predictor network)
+            part
+        """
+        # Now for the decoder, i.e., the prediction network
+        
+
+        blank_id = self.decoder.blank_id
+        sos_y = add_sos(y, sos_id=blank_id)
+
+        # sos_y_padded: [B, S + 1], start with SOS.
+        sos_y_padded = sos_y.pad(mode="constant", padding_value=blank_id)
+
+        # decoder_out: [B, S + 1, decoder_dim]
+        decoder_out = self.decoder(sos_y_padded)
+
+        # Note: y does not start with SOS
+        # y_padded : [B, S]
+        y_padded = y.pad(mode="constant", padding_value=0)
+
+        y_padded = y_padded.to(torch.int64)
+        boundary = torch.zeros(
+            (encoder_out.size(0), 4),
+            dtype=torch.int64,
+            device=encoder_out.device,
+        )
+        boundary[:, 2] = y_lens
+        boundary[:, 3] = encoder_out_lens
+        lm = self.simple_lm_proj(decoder_out)
+        am = self.simple_am_proj(encoder_out)
+
+        # if self.training and random.random() < 0.25:
+        #    lm = penalize_abs_values_gt(lm, 100.0, 1.0e-04)
+        # if self.training and random.random() < 0.25:
+        #    am = penalize_abs_values_gt(am, 30.0, 1.0e-04)
+
+        with torch.cuda.amp.autocast(enabled=False):
+          simple_loss, (px_grad, py_grad) = k2.rnnt_loss_smoothed(
+              lm=lm.float(),
+              am=am.float(),
+              symbols=y_padded,
+              termination_symbol=blank_id,
+              lm_only_scale=lm_scale,
+              am_only_scale=am_scale,
+              boundary=boundary,
+              reduction="sum",
+              return_grad=True,
+          )
+          
+
+        # am_pruned : [B, T, prune_range, encoder_dim]
+        # lm_pruned : [B, T, prune_range, decoder_dim]
+       
+        # ranges : [B, T, prune_range]
+        ranges = k2.get_rnnt_prune_ranges(
+            px_grad=px_grad,
+            py_grad=py_grad,
+            boundary=boundary,
+            s_range=prune_range,
+        )
+        am_pruned, lm_pruned = k2.do_rnnt_pruning(
+            am=self.joiner.encoder_proj(encoder_out),
+            lm=self.joiner.decoder_proj(decoder_out),
+            ranges=ranges,
+        )
+          # project_input=False since we applied the decoder's input projections
+          # prior to do_rnnt_pruning (this is an optimization for speed).
+        logits = self.joiner(am_pruned, lm_pruned, project_input=False)
+        with torch.cuda.amp.autocast(enabled=False):
+          pruned_loss = k2.rnnt_loss_pruned(
+              logits=logits.float(),
+              symbols=y_padded,
+              ranges=ranges,
+              termination_symbol=blank_id,
+              boundary=boundary,
+              reduction="sum",
+              use_hat_loss=self.use_hat,
+          )
+         # logits : [B, T, prune_range, vocab_size]
+        return simple_loss, pruned_loss
+
+    def forward(
+        self,
+        x: torch.Tensor,
+        x_lens: torch.Tensor,
+        y: k2.RaggedTensor,
+        st_y: k2.RaggedTensor,
+        prune_range: int = 5,
+        st_prune_range: int =10,
+        am_scale: float = 0.0,
+        lm_scale: float = 0.0,
+        use_st_cr_ctc: bool = False,
+        use_asr_cr_ctc: bool = False,
+        use_spec_aug: bool = False,
+        spec_augment: Optional[SpecAugment] = None,
+        supervision_segments: Optional[torch.Tensor] = None,
+        time_warp_factor: Optional[int] = 80,
+    ) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor]:
+        """
+        Args:
+          x:
+            A 3-D tensor of shape (N, T, C).
+          x_lens:
+            A 1-D tensor of shape (N,). It contains the number of frames in `x`
+            before padding.
+          y:
+            A ragged tensor with 2 axes [utt][label]. It contains labels of each
+            utterance.
+          prune_range:
+            The prune range for rnnt loss, it means how many symbols(context)
+            we are considering for each frame to compute the loss.
+          am_scale:
+            The scale to smooth the loss with am (output of encoder network)
+            part
+          lm_scale:
+            The scale to smooth the loss with lm (output of predictor network)
+            part
+          use_cr_ctc:
+            Whether use consistency-regularized CTC.
+          use_spec_aug:
+            Whether apply spec-augment manually, used only if use_cr_ctc is True.
+          spec_augment:
+            The SpecAugment instance that returns time masks,
+            used only if use_cr_ctc is True.
+          supervision_segments:
+            An int tensor of shape ``(S, 3)``. ``S`` is the number of
+            supervision segments that exist in ``features``.
+            Used only if use_cr_ctc is True.
+          time_warp_factor:
+            Parameter for the time warping; larger values mean more warping.
+            Set to ``None``, or less than ``1``, to disable.
+            Used only if use_cr_ctc is True.
+            
+        Returns:
+          Return the transducer losses and CTC loss,
+          in form of (simple_loss, pruned_loss, ctc_loss)
+
+        Note:
+           Regarding am_scale & lm_scale, it will make the loss-function one of
+           the form:
+              lm_scale * lm_probs + am_scale * am_probs +
+              (1-lm_scale-am_scale) * combined_probs
+        """
+        assert x.ndim == 3, x.shape
+        assert x_lens.ndim == 1, x_lens.shape
+        assert y.num_axes == 2, y.num_axes
+        assert st_y.num_axes == 2, st_y.num_axes
+        assert x.size(0) == x_lens.size(0) == y.dim0, (x.shape, x_lens.shape, y.dim0)
+        device = x.device
+        if use_st_cr_ctc or use_asr_cr_ctc:
+          assert self.use_ctc or self.use_st_ctc
+          if use_spec_aug:
+              assert spec_augment is not None and spec_augment.time_warp_factor < 1
+              # Apply time warping before input duplicating
+              assert supervision_segments is not None
+              x = time_warp(
+                  x,
+                  time_warp_factor=time_warp_factor,
+                  supervision_segments=supervision_segments,
+              )
+              # Independently apply frequency masking and time masking to the two copies
+              x = spec_augment(x.repeat(2, 1, 1))
+          else:
+              x = x.repeat(2, 1, 1)
+          x_lens = x_lens.repeat(2)
+          y = k2.ragged.cat([y, y], axis=0)
+          if self.st_joiner != None and self.use_st_ctc:
+            st_y = k2.ragged.cat([st_y, st_y], axis=0)
+        # Compute encoder outputs
+        encoder_out, encoder_out_lens, st_encoder_out, st_encoder_out_lens = self.forward_encoder(x, x_lens)
+
+        row_splits = y.shape.row_splits(1)
+        y_lens = row_splits[1:] - row_splits[:-1]
+        st_row_splits = st_y.shape.row_splits(1)
+        st_y_lens = st_row_splits[1:] - st_row_splits[:-1]
+
+        if self.use_transducer:
+            # Compute transducer loss
+            if self.st_joiner != None:
+              simple_loss, st_simple_loss, pruned_loss, st_pruned_loss = self.forward_st_transducer(
+                  st_encoder_out=st_encoder_out,
+                  st_encoder_out_lens=st_encoder_out_lens,
+                  encoder_out=encoder_out,
+                  encoder_out_lens=encoder_out_lens,
+                  y=y.to(x.device),
+                  y_lens=y_lens,
+                  st_y=st_y.to(x.device),
+                  st_y_lens=st_y_lens,
+                  prune_range=st_prune_range,
+                  am_scale=am_scale,
+                  lm_scale=lm_scale,
+              )
+              if use_asr_cr_ctc:
+                simple_loss = simple_loss * 0.5
+                pruned_loss = pruned_loss * 0.5
+              if use_st_cr_ctc:
+                st_simple_loss = st_simple_loss * 0.5
+                st_pruned_loss = st_pruned_loss * 0.5
+            else:
+                simple_loss, pruned_loss = self.forward_transducer(
+                  encoder_out=encoder_out,
+                  encoder_out_lens=encoder_out_lens,
+                  y=y.to(x.device),
+                  y_lens=y_lens,
+                  prune_range=prune_range,
+                  am_scale=am_scale,
+                  lm_scale=lm_scale,
+              )
+                if use_asr_cr_ctc:
+                  simple_loss = simple_loss * 0.5
+                  pruned_loss = pruned_loss * 0.5
+                st_simple_loss, st_pruned_loss = torch.empty(0), torch.empty(0)
+        else:
+            simple_loss = torch.empty(0)
+            pruned_loss = torch.empty(0)
+
+        if self.use_ctc:
+            # Compute CTC loss
+            targets = y.values
+            if not use_asr_cr_ctc:
+                ctc_loss = self.forward_ctc(
+                    encoder_out=encoder_out,
+                    encoder_out_lens=encoder_out_lens,
+                    targets=targets,
+                    target_lengths=y_lens,
+                )
+                cr_loss = torch.empty(0)
+            else:
+                ctc_loss, cr_loss = self.forward_cr_ctc(
+                    encoder_out=encoder_out,
+                    encoder_out_lens=encoder_out_lens,
+                    targets=targets,
+                    target_lengths=y_lens,
+                )
+                ctc_loss = ctc_loss * 0.5
+                cr_loss = cr_loss * 0.5
+        else:
+            ctc_loss = torch.empty(0)
+            cr_loss = torch.empty(0)
+        if self.use_st_ctc:
+            # Compute CTC loss
+            st_targets = st_y.values
+            if not use_st_cr_ctc:
+                st_ctc_loss = self.forward_st_ctc(
+                    st_encoder_out=st_encoder_out,
+                    st_encoder_out_lens=st_encoder_out_lens,
+                    targets=st_targets,
+                    target_lengths=st_y_lens,
+                )
+                st_cr_loss = torch.empty(0)
+            else:
+                st_ctc_loss, st_cr_loss = self.forward_st_cr_ctc(
+                    st_encoder_out=st_encoder_out,
+                    st_encoder_out_lens=st_encoder_out_lens,
+                    st_targets=st_targets,
+                    st_target_lengths=st_y_lens,
+                    # encoder_out=encoder_out,
+                    # encoder_out_lens=encoder_out_lens,
+                    # targets=targets,
+                    # target_lengths=y_lens,
+                )
+                st_ctc_loss = st_ctc_loss * 0.5
+                st_cr_loss = st_cr_loss * 0.5
+        else:
+            st_ctc_loss = torch.empty(0)
+            st_cr_loss = torch.empty(0)
+
+        return simple_loss, st_simple_loss, pruned_loss, st_pruned_loss, ctc_loss, st_ctc_loss, cr_loss, st_cr_loss

egs/multi_conv_zh_es_ta/ST/hent_srt/optim.py

@@ -0,0 +1 @@
+../zipformer_multijoiner_st/optim.py

egs/multi_conv_zh_es_ta/ST/hent_srt/profile.py

@@ -0,0 +1 @@
+../zipformer_multijoiner_st/profile.py

egs/multi_conv_zh_es_ta/ST/hent_srt/scaling.py

@@ -0,0 +1 @@
+../zipformer_multijoiner_st/scaling.py

egs/multi_conv_zh_es_ta/ST/hent_srt/scaling_converter.py

@@ -0,0 +1 @@
+../zipformer_multijoiner_st/scaling_converter.py

egs/multi_conv_zh_es_ta/ST/hent_srt/streaming_beam_search.py

@@ -0,0 +1,244 @@
+# Copyright      2025  Johns Hopkins University (author: Amir Hussein)
+# Copyright    2022  Xiaomi Corp.        (authors: Wei Kang)
+#
+# See ../../../../LICENSE for clarification regarding multiple authors
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import warnings
+from typing import List
+
+import k2
+import torch
+import torch.nn as nn
+from beam_search import Hypothesis, HypothesisList, get_hyps_shape
+from decode_stream import DecodeStream
+
+from icefall.decode import one_best_decoding
+from icefall.utils import get_texts
+
+
+def greedy_search_st(
+    model: nn.Module,
+    encoder_out: torch.Tensor,
+    encoder_out_st: torch.Tensor,
+    max_sym_per_frame: int,
+    streams: List[DecodeStream],
+    st_blank_penalty: float = 0.0,
+) -> None:
+    """Greedy search in batch mode. It hardcodes --max-sym-per-frame=1.
+
+    Args:
+      model:
+        The transducer model.
+      encoder_out:
+        Output from the encoder. Its shape is (N, T, C), where N >= 1.
+      streams:
+        A list of Stream objects.
+    """
+    assert len(streams) == encoder_out_st.size(0)
+    assert encoder_out_st.ndim == 3
+    # ST
+    blank_id_st = model.st_decoder.blank_id
+    context_size_st = model.st_decoder.context_size
+    unk_id_st = getattr(model, "unk_id", blank_id_st)
+    device = model.device
+    T = encoder_out_st.size(1)
+    # ASR
+    blank_id = model.decoder.blank_id
+    context_size = model.decoder.context_size
+    unk_id = getattr(model, "unk_id", blank_id_st)
+
+    #ST
+    
+    decoder_input_st = torch.tensor(
+        [stream.hyp_st[-context_size_st:] for stream in streams],
+        device=device,
+        dtype=torch.int64,
+    )
+    # decoder_out is of shape (N, 1, decoder_out_dim)
+    decoder_out_st = model.st_decoder(decoder_input_st, need_pad=False)
+    decoder_out_st = model.st_joiner.decoder_proj(decoder_out_st)
+    
+    # ASR
+    decoder_input = torch.tensor(
+        [stream.hyp_asr[-context_size:] for stream in streams],
+        device=device,
+        dtype=torch.int64,
+    )
+    decoder_out = model.decoder(decoder_input, need_pad=False)
+    decoder_out = model.joiner.decoder_proj(decoder_out)
+    # Maximum symbols per utterance.
+    max_sym_per_utt = 10000
+    # symbols per frame
+    sym_per_frame = 0
+
+    # symbols per utterance decoded so far
+    sym_per_utt = 0
+    t = 0
+    # for t in range(T):
+    while t < T and sym_per_utt < max_sym_per_utt:
+        if sym_per_frame >= max_sym_per_frame:
+            sym_per_frame = 0
+            t += 1
+            continue
+        # current_encoder_out's shape: (batch_size, 1, encoder_out_dim)
+        # current_encoder_out_st = encoder_out_st[:, t : t + 1, :] # noqa
+        current_encoder_out_st = encoder_out_st[:, t : t + 1, :].unsqueeze(2)
+
+        st_logits = model.st_joiner(
+            current_encoder_out_st,
+            decoder_out_st.unsqueeze(1),
+            project_input=False,
+        )
+        # logits'shape (batch_size,  vocab_size)
+        st_logits = st_logits.squeeze(1).squeeze(1)
+        if st_blank_penalty != 0.0:
+            st_logits[:, 0] -= st_blank_penalty
+
+        assert st_logits.ndim == 2, st_logits.shape
+        y_st = st_logits.argmax(dim=1).tolist()
+        for i, v in enumerate(y_st):
+            if v not in (blank_id_st, unk_id_st):
+                streams[i].hyp_st.append(v)
+               
+                # update decoder output
+                # decoder_input_st = torch.tensor(
+                #     [stream.hyp_st[-context_size_st:].reshape(
+                # 1, context_size_st) for stream in streams],
+                #     device=device,
+                #     dtype=torch.int64,
+                # )
+                decoder_input_st = torch.stack([
+                torch.tensor(stream.hyp_st[-context_size_st:], device=device, dtype=torch.int64)
+                for stream in streams]).reshape(len(streams), context_size_st)
+                
+                decoder_out_st = model.st_decoder(
+                    decoder_input_st,
+                    need_pad=False,
+                )
+                decoder_out_st = model.st_joiner.decoder_proj(decoder_out_st)
+                sym_per_utt += 1
+                sym_per_frame += 1
+            else:
+                sym_per_frame = 0
+                t += 1
+
+
+def modified_beam_search(
+    model: nn.Module,
+    encoder_out: torch.Tensor,
+    streams: List[DecodeStream],
+    num_active_paths: int = 4,
+    blank_penalty: float = 0.0,
+) -> None:
+    """Beam search in batch mode with --max-sym-per-frame=1 being hardcoded.
+
+    Args:
+      model:
+        The RNN-T model.
+      encoder_out:
+        A 3-D tensor of shape (N, T, encoder_out_dim) containing the output of
+        the encoder model.
+      streams:
+        A list of stream objects.
+      num_active_paths:
+        Number of active paths during the beam search.
+    """
+    assert encoder_out.ndim == 3, encoder_out.shape
+    assert len(streams) == encoder_out.size(0)
+
+    blank_id = model.decoder.blank_id
+    context_size = model.decoder.context_size
+    device = next(model.parameters()).device
+    batch_size = len(streams)
+    T = encoder_out.size(1)
+
+    B = [stream.hyps for stream in streams]
+
+    for t in range(T):
+        current_encoder_out = encoder_out[:, t].unsqueeze(1).unsqueeze(1)
+        # current_encoder_out's shape: (batch_size, 1, 1, encoder_out_dim)
+
+        hyps_shape = get_hyps_shape(B).to(device)
+
+        A = [list(b) for b in B]
+        B = [HypothesisList() for _ in range(batch_size)]
+
+        ys_log_probs = torch.stack(
+            [hyp.log_prob.reshape(1) for hyps in A for hyp in hyps], dim=0
+        )  # (num_hyps, 1)
+
+        decoder_input = torch.tensor(
+            [hyp.ys[-context_size:] for hyps in A for hyp in hyps],
+            device=device,
+            dtype=torch.int64,
+        )  # (num_hyps, context_size)
+
+        decoder_out = model.decoder(decoder_input, need_pad=False).unsqueeze(1)
+        decoder_out = model.joiner.decoder_proj(decoder_out)
+        # decoder_out is of shape (num_hyps, 1, 1, decoder_output_dim)
+
+        # Note: For torch 1.7.1 and below, it requires a torch.int64 tensor
+        # as index, so we use `to(torch.int64)` below.
+        current_encoder_out = torch.index_select(
+            current_encoder_out,
+            dim=0,
+            index=hyps_shape.row_ids(1).to(torch.int64),
+        )  # (num_hyps, encoder_out_dim)
+
+        logits = model.joiner(current_encoder_out, decoder_out, project_input=False)
+        # logits is of shape (num_hyps, 1, 1, vocab_size)
+
+        logits = logits.squeeze(1).squeeze(1)
+
+        if blank_penalty != 0.0:
+            logits[:, 0] -= blank_penalty
+
+        log_probs = logits.log_softmax(dim=-1)  # (num_hyps, vocab_size)
+
+        log_probs.add_(ys_log_probs)
+
+        vocab_size = log_probs.size(-1)
+
+        log_probs = log_probs.reshape(-1)
+
+        row_splits = hyps_shape.row_splits(1) * vocab_size
+        log_probs_shape = k2.ragged.create_ragged_shape2(
+            row_splits=row_splits, cached_tot_size=log_probs.numel()
+        )
+        ragged_log_probs = k2.RaggedTensor(shape=log_probs_shape, value=log_probs)
+
+        for i in range(batch_size):
+            topk_log_probs, topk_indexes = ragged_log_probs[i].topk(num_active_paths)
+
+            with warnings.catch_warnings():
+                warnings.simplefilter("ignore")
+                topk_hyp_indexes = (topk_indexes // vocab_size).tolist()
+                topk_token_indexes = (topk_indexes % vocab_size).tolist()
+
+            for k in range(len(topk_hyp_indexes)):
+                hyp_idx = topk_hyp_indexes[k]
+                hyp = A[i][hyp_idx]
+
+                new_ys = hyp.ys[:]
+                new_token = topk_token_indexes[k]
+                if new_token != blank_id:
+                    new_ys.append(new_token)
+
+                new_log_prob = topk_log_probs[k]
+                new_hyp = Hypothesis(ys=new_ys, log_prob=new_log_prob)
+                B[i].add(new_hyp)
+
+    for i in range(batch_size):
+        streams[i].hyps = B[i]

egs/multi_conv_zh_es_ta/ST/hent_srt/subsampling.py

@@ -0,0 +1 @@
+../zipformer_multijoiner_st/subsampling.py

egs/multi_conv_zh_es_ta/ST/local/cer.py

@@ -0,0 +1,73 @@
+import argparse 
+import jiwer
+import os 
+import re
+
+def get_args():
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "--dec-file",
+        type=str,
+        help="file with decoded text"
+    )
+
+    return parser
+
+
+def contains_chinese(text):
+    """
+    Check if the given text contains any Chinese characters.
+
+    Args:
+        text (str): The input string.
+
+    Returns:
+        bool: True if the string contains at least one Chinese character, False otherwise.
+    """
+    chinese_char_pattern = re.compile(r'[\u4e00-\u9fff]')
+    return bool(chinese_char_pattern.search(text))
+
+def cer_(file):
+    hyp = []
+    ref = []
+    cer_results = 0
+    ref_lens = 0
+    with open(file, 'r', encoding='utf-8') as dec:
+        
+        for line in dec:
+            id, target = line.split('\t')
+            id = id[0:-1]
+            target, txt = target.split("=")
+            
+            if target == 'ref':
+                words = txt.strip().strip('[]').split(', ')
+                word_list = [word.strip("'") for word in words]
+                # if contains_chinese(" ".join(word_list)):
+                #     word_list = [" ".join(re.findall(r".",word.strip("'"))) for word in words]
+                ref.append("".join(word_list))
+            elif target == 'hyp':
+                words = txt.strip().strip('[]').split(', ')
+               
+                word_list = [word.strip("'") for word in words]
+                # if contains_chinese(" ".join(word_list)):
+                    
+                #     word_list = ["".join(re.findall(r".",word.strip("'"))) for word in words]
+                hyp.append("".join(word_list))
+        for h, r in zip(hyp, ref):
+            if r:
+                cer_results += (jiwer.cer(r, h)*len(r))
+                
+                ref_lens += len(r)
+    #print(os.path.basename(file))
+    print(cer_results / ref_lens)
+
+        
+
+
+def main():
+    parse = get_args()
+    args = parse.parse_args() 
+    cer_(args.dec_file)
+    
+if __name__ == "__main__":
+    main()

egs/multi_conv_zh_es_ta/ST/local/compute_fbank_gpu.py

@@ -0,0 +1,175 @@
+#!/usr/bin/env python3
+# Johns Hopkins University  (authors: Amir Hussein)
+#
+# See ../../../../LICENSE for clarification regarding multiple authors
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+
+"""
+This file computes fbank features.
+It looks for manifests in the directory data/manifests.
+
+The generated fbank features are saved in data/fbank.
+"""
+
+import logging
+import os
+from pathlib import Path
+import argparse
+
+import torch
+from lhotse import CutSet, Fbank, FbankConfig, LilcomChunkyWriter
+from lhotse.recipes.utils import read_manifests_if_cached
+
+from icefall.utils import get_executor
+
+from lhotse.features.kaldifeat import (
+    KaldifeatFbank,
+    KaldifeatFbankConfig,
+    KaldifeatFrameOptions,
+    KaldifeatMelOptions,
+)
+
+# Torch's multithreaded behavior needs to be disabled or
+# it wastes a lot of CPU and slow things down.
+# Do this outside of main() in case it needs to take effect
+# even when we are not invoking the main (e.g. when spawning subprocesses)
+
+def get_args():
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "--num-splits",
+        type=int,
+        default=5,
+        help="Number of splits for the train set.",
+    )
+    parser.add_argument(
+        "--start",
+        type=int,
+        default=0,
+        help="Start index of the train set split.",
+    )
+    parser.add_argument(
+        "--stop",
+        type=int,
+        default=-1,
+        help="Stop index of the train set split.",
+    )
+    parser.add_argument(
+        "--test",
+        action="store_true",
+        help="If set, only compute features for the dev and val set.",
+    )
+    parser.add_argument(
+        "--datadir",
+        type=str,
+        help="Manifests datadir",
+    )
+
+    return parser.parse_args()
+
+
+def compute_fbank_gpu(args):
+    src_dir = Path(args.datadir+"/manifests")
+    output_dir = Path(args.datadir+"/fbank")
+    num_jobs = min(os.cpu_count(),10)
+    num_mel_bins = 80
+    sampling_rate = 16000
+    sr = 16000
+
+    logging.info(f"Cpus {num_jobs}")
+
+
+    if args.test:
+        dataset_parts = (
+            "hkust_test",
+            "iwslt_ta_test",
+            "fisher-sp_test",
+            "dev")
+    else:
+        dataset_parts = (
+            "train")
+    prefix = "cts"
+    suffix = "jsonl.gz"
+    manifests = read_manifests_if_cached(
+        prefix=prefix, dataset_parts=dataset_parts, output_dir=src_dir,suffix=suffix,
+    )
+    assert manifests is not None
+
+    extractor = KaldifeatFbank(
+        KaldifeatFbankConfig(
+            frame_opts=KaldifeatFrameOptions(sampling_rate=sampling_rate),
+            mel_opts=KaldifeatMelOptions(num_bins=num_mel_bins),
+            device="cuda",
+        )
+    )
+
+    for partition, m in manifests.items():
+        cuts_filename = f"{prefix}_cuts_{partition}.{suffix}"
+        if (output_dir / f"{cuts_filename}").is_file():
+            logging.info(f"{partition} already exists - skipping.")
+            continue
+        logging.info(f"Processing {partition}")
+        cut_set = CutSet.from_manifests(
+            recordings=m["recordings"],
+            supervisions=m["supervisions"],
+        )
+
+        logging.info("About to split cuts into smaller chunks.")
+        if sr != None:
+            logging.info(f"Resampling to {sr}")
+            cut_set = cut_set.resample(sr)
+
+        cut_set = cut_set.trim_to_supervisions(
+                    keep_overlapping=False, 
+                    keep_all_channels=False)
+        cut_set = cut_set.filter(lambda c: c.duration >= .2 and c.duration <= 30)
+        if "train" in partition:
+            cut_set = (
+                cut_set
+                + cut_set.perturb_speed(0.9)
+                + cut_set.perturb_speed(1.1)
+            )
+            cut_set = cut_set.compute_and_store_features_batch(
+                extractor=extractor,
+                storage_path=f"{output_dir}/{prefix}_feats_{partition}",
+                manifest_path=f"{src_dir}/{cuts_filename}",
+                batch_duration=2000,
+                num_workers=num_jobs,
+                storage_type=LilcomChunkyWriter,
+                overwrite=True,
+                )
+            cut_set.to_file(output_dir / f"cuts_{partition}.jsonl.gz")
+        else:
+            logging.info(f"Processing {partition}")
+            cut_set = cut_set.compute_and_store_features_batch(
+                extractor=extractor,
+                storage_path=f"{output_dir}/{prefix}_feats_{partition}",
+                manifest_path=f"{src_dir}/{cuts_filename}",
+                batch_duration=2000,
+                num_workers=num_jobs,
+                storage_type=LilcomChunkyWriter,
+                overwrite=True,
+                )
+            cut_set.to_file(output_dir / f"cuts_{partition}.jsonl.gz")
+
+if __name__ == "__main__":
+    formatter = (
+        "%(asctime)s %(levelname)s [%(filename)s:%(lineno)d] %(message)s"
+    )
+
+    logging.basicConfig(format=formatter, level=logging.INFO)
+    args = get_args()
+
+    compute_fbank_gpu(args)

egs/multi_conv_zh_es_ta/ST/local/compute_fbank_musan.py

@@ -0,0 +1 @@
+../../../librispeech/ASR/local/compute_fbank_musan.py

egs/multi_conv_zh_es_ta/ST/local/cuts_validate.py

@@ -0,0 +1,97 @@
+#!/usr/bin/python
+
+from lhotse import RecordingSet, SupervisionSet, CutSet
+import argparse
+import logging
+from lhotse.qa import fix_manifests, validate_recordings_and_supervisions
+import pdb
+
+
+
+def get_parser():
+    parser = argparse.ArgumentParser(
+        formatter_class=argparse.ArgumentDefaultsHelpFormatter
+    )
+
+    parser.add_argument(
+        "--sup",
+        type=str,
+        default="",
+        help="Supervisions file",
+    )
+
+    parser.add_argument(
+        "--rec",
+        type=str,
+        default="",
+        help="Recordings file",
+    )
+    parser.add_argument(
+        "--cut",
+        type=str,
+        default="",
+        help="Cutset file",
+    )
+    parser.add_argument(
+        "--savecut",
+        type=str,
+        default="",
+        help="name of the cutset to be saved",
+    )
+
+    return parser
+
+
+def valid_asr(cut):
+    tol = 2e-3
+    i=0
+    total_dur = 0
+    for c in cut:
+        if c.supervisions != []:
+            if c.supervisions[0].end > c.duration + tol:
+
+                logging.info(f"Supervision beyond the cut. Cut number: {i}")
+                total_dur += c.duration
+                logging.info(f"id: {c.id}, sup_end: {c.supervisions[0].end},  dur: {c.duration}, source {c.recording.sources[0].source}")
+            elif c.supervisions[0].start < -tol:
+                logging.info(f"Supervision starts before the cut. Cut number: {i}")
+                logging.info(f"id: {c.id}, sup_start: {c.supervisions[0].start},  dur: {c.duration}, source {c.recording.sources[0].source}")
+            else:
+                continue
+        else:
+            logging.info("Empty supervision")
+            logging.info(f"id: {c.id}")
+        i += 1
+    logging.info(f"filtered duration: {total_dur}")
+     
+
+def main():
+
+    parser = get_parser()
+    args = parser.parse_args()
+    if args.cut != "":
+        cuts = CutSet.from_file(args.cut)
+    else:
+        recordings = RecordingSet.from_file(args.rec)
+        supervisions = SupervisionSet.from_file(args.sup)
+       # breakpoint()
+        logging.info("Example from supervisions:")
+        logging.info(supervisions[0])
+        logging.info("Example from recordings")
+        logging.info("Fixing manifests")
+        recordings, supervisions = fix_manifests(recordings, supervisions)
+        logging.info("Validating manifests")
+        validate_recordings_and_supervisions(recordings, supervisions)
+    
+        cuts = CutSet.from_manifests(recordings= recordings, supervisions=supervisions,)
+    cuts = cuts.trim_to_supervisions(keep_overlapping=False, keep_all_channels=False)
+    cuts.describe()
+    logging.info("Example from cut:")
+    logging.info(cuts[100])
+    logging.info("Validating manifests for ASR")
+    valid_asr(cuts)
+    if args.savecut != "":
+        cuts.to_file(args.savecut)
+
+if __name__ == "__main__":
+    main()

egs/multi_conv_zh_es_ta/ST/local/prepare_st_transcripts.py

@@ -0,0 +1,53 @@
+# Copyright 2023 Johns Hopkins University  (Amir Hussein)
+
+#!/usr/bin/python
+"""
+This script prepares transcript_words.txt from cutset
+"""
+
+from lhotse import CutSet
+import argparse
+import logging
+import pdb
+from pathlib import Path
+import os
+
+
+def get_parser():
+    parser = argparse.ArgumentParser(
+        formatter_class=argparse.ArgumentDefaultsHelpFormatter
+    )
+    parser.add_argument(
+        "--cut",
+        type=str,
+        default="",
+        help="Cutset file",
+    )
+    parser.add_argument(
+        "--langdir",
+        type=str,
+        default="",
+        help="name of the lang-dir",
+    )
+    return parser
+     
+
+def main():
+
+    parser = get_parser()
+    args = parser.parse_args()
+
+    logging.info("Reading the cuts")
+    cuts = CutSet.from_file(args.cut)
+    langdir = Path(args.langdir)
+    
+    if not os.path.exists(langdir):
+        os.makedirs(langdir)
+    
+    with open(langdir / "st_words.txt", 'w') as txt:
+        for c in cuts:
+            text = c.supervisions[0].custom['translated_text']['en']
+            txt.write(text + '\n')
+
+if __name__ == "__main__":
+    main()

egs/multi_conv_zh_es_ta/ST/local/prepare_transcripts.py

@@ -0,0 +1,54 @@
+# Copyright 2023 Johns Hopkins University  (Amir Hussein)
+
+#!/usr/bin/python
+"""
+This script prepares transcript_words.txt from cutset
+"""
+
+from lhotse import CutSet
+import argparse
+import logging
+import pdb
+from pathlib import Path
+import os
+
+
+def get_parser():
+    parser = argparse.ArgumentParser(
+        formatter_class=argparse.ArgumentDefaultsHelpFormatter
+    )
+    parser.add_argument(
+        "--cut",
+        type=str,
+        default="",
+        help="Cutset file",
+    )
+    parser.add_argument(
+        "--langdir",
+        type=str,
+        default="",
+        help="name of the lang-dir",
+    )
+    return parser
+     
+
+def main():
+
+    parser = get_parser()
+    args = parser.parse_args()
+
+    logging.info("Reading the cuts")
+    cuts = CutSet.from_file(args.cut)
+    langdir = Path(args.langdir)
+    
+    if not os.path.exists(langdir):
+        os.makedirs(langdir)
+    
+    with open(langdir / "transcript_words.txt", 'w') as txt:
+        for c in cuts:
+            #breakpoint()
+            text = c.supervisions[0].text
+            txt.write(text + '\n')
+
+if __name__ == "__main__":
+    main()

egs/multi_conv_zh_es_ta/ST/local/train_bpe_model.py

@@ -0,0 +1 @@
+../../../librispeech/ASR/local/train_bpe_model.py

egs/multi_conv_zh_es_ta/ST/prepare.sh

@@ -0,0 +1,175 @@
+#!/usr/bin/env bash
+# Copyright 2023 Johns Hopkins University  (Amir Hussein)
+# Apache 2.0  (http://www.apache.org/licenses/LICENSE-2.0)
+
+set -eou pipefail
+
+nj=20
+stage=0
+stop_stage=7
+
+# We assume dl_dir (download dir) contains the following
+# directories and files.
+#
+#  - $dl_dir/cts
+#
+#      You can download the data from
+#
+#
+#  - $dl_dir/musan
+#      This directory contains the following directories downloaded from
+#       http://www.openslr.org/17/
+#
+#     - music
+#     - noise
+#     - speech
+#
+
+dl_dir=cts
+. shared/parse_options.sh || exit 1
+
+# vocab size for sentence piece models.
+# It will generate data/lang_bpe_xxx,
+# data/lang_bpe_yyy if the array contains xxx, yyy
+vocab_sizes=(
+  5000
+)
+st_vocab_sizes=(
+  4000
+)
+
+# All files generated by this script are saved in "data".
+# You can safely remove "data" and rerun this script to regenerate it.
+mkdir -p data
+
+log() {
+  # This function is from espnet
+  local fname=${BASH_SOURCE[1]##*/}
+  echo -e "$(date '+%Y-%m-%d %H:%M:%S') (${fname}:${BASH_LINENO[0]}:${FUNCNAME[1]}) $*"
+}
+
+log "dl_dir: $dl_dir"
+
+if [ $stage -le 0 ] && [ $stop_stage -ge 0 ]; then
+  log "Stage 0: Download data"
+
+  # Download callhome_spanish, fisher_spanish iwslt22_ta and HKUST from LDC 
+  # 
+  # you can create a symlink
+  #
+  #   ln -sfv /path/to/data $dl_dir/data
+
+  # If you have pre-downloaded it to /path/to/musan,
+  # you can create a symlink
+  #
+  #   ln -sfv /path/to/musan $dl_dir/
+  #
+  if [ ! -d $dl_dir/musan ]; then
+    lhotse download musan $dl_dir
+  fi
+fi
+
+fbank=data/fbank
+manifests=data/manifests
+mkdir -p $manifests
+sets="hkust iwslt-ta callhome-sp fisher-sp"
+if [ $stage -le 0 ] && [ $stop_stage -ge 0 ]; then
+  log "Stage 0: Prepare telephone manifest"
+  # We assume that you have downloaded callhome_spanish, fisher_spanish iwslt22_ta and hkust to $dl_dir/
+  for set in $sets; do
+    log "Prepare $set manifests"
+    if [[ "$set" == "iwslt-ta" ]]; then
+      if [ ! -d "iwslt22-dialect" ]; then
+        echo "Splits directory (iwslt22-dialect) does not exist"
+        echo "Run: git clone https://github.com/kevinduh/iwslt22-dialect"
+        exit 1
+      else
+        lhotse prepare "$set" "$dl_dir/$set" iwslt22-dialect "$manifests"
+      fi
+    else
+      lhotse prepare "$set" "$dl_dir/$set" "$manifests"
+      # validate recordings and supervisions
+    fi
+    # python local/cuts_validate.py \
+    #     --sup "${manifests}/supervisions.jsonl.gz" \
+    #     --rec "${manifests}/recordings.jsonl.gz" \
+    #     --savecut "${manifests}/cuts_${set}.jsonl.gz"
+  done
+fi
+
+if [ $stage -le 1 ] && [ $stop_stage -ge 1 ]; then
+  if [ ! -f ${manifests}/cut_train.jsonl.gz ]; then
+    log "Combining conversational data to create train, dev sets"
+    # combine train
+    lhotse combine $manifests/iwslt-ta_supervisions_train.jsonl.gz  $manifests/hkust_supervisions_train.jsonl.gz $manifests/fisher-sp_supervisions_train.jsonl.gz ${manifests}/cts_supervisions_train.jsonl.gz
+    
+    lhotse combine $manifests/iwslt-ta_recordings_train.jsonl.gz  $manifests/hkust_recordings_train.jsonl.gz  $manifests/fisher-sp_recordings_train.jsonl.gz ${manifests}/cts_recordings_train.jsonl.gz
+    
+    # python local/cuts_validate.py --sup $manifests/cts_supervisions_train.jsonl.gz --rec ${manifests}/cts_recordings_train.jsonl.gz
+
+    # combine dev
+    lhotse combine $manifests/iwslt-ta_supervisions_dev1.jsonl.gz  $manifests/hkust_supervisions_dev1.jsonl.gz $manifests/fisher-sp_supervisions_dev.jsonl.gz  ${manifests}/cts_supervisions_dev.jsonl.gz
+        
+    lhotse combine $manifests/iwslt-ta_recordings_dev1.jsonl.gz $manifests/fisher-spanish_recordings_dev.jsonl.gz $manifests/hkust_recordings_dev1.jsonl.gz $manifests/fisher-sp_recordings_dev.jsonl.gz ${manifests}/cts_recordings_dev.jsonl.gz
+
+    # python local/cuts_validate.py --sup ${manifests}/cts_supervisions_dev.jsonl.gz --rec ${manifests}/cts_recordings_dev.jsonl.gz
+  fi
+fi
+
+if [ $stage -le 2 ] && [ $stop_stage -ge 2 ]; then
+  log "Stage 2: Prepare musan manifest"
+  # We assume that you have downloaded the musan corpus
+  # to data/musan
+  if [ ! -f ${manifests}/musan_recordings_speech.jsonl.gz ]; then
+    mkdir -p $manifests
+    lhotse prepare musan $dl_dir/musan $manifests
+  fi
+fi
+
+if [ $stage -le 3 ] && [ $stop_stage -ge 3 ]; then
+  log "Stage 3: Compute fbank features"
+  mkdir -p ${fbank}
+  ./local/compute_fbank_gpu.py
+  ./local/compute_fbank_gpu.py --test
+fi
+
+if [ $stage -le 4 ] && [ $stop_stage -ge 4 ]; then
+  log "Stage 4: Compute fbank for musan"
+  ./local/compute_fbank_musan.py
+fi
+
+if [ $stage -le 6 ] && [ $stop_stage -ge 6 ]; then
+  log "Stage 6: Prepare BPE based lang"
+  for vocab_size in ${vocab_sizes[@]}; do
+    lang_dir=data/lang_bpe_${vocab_size}
+    mkdir -p ${lang_dir}
+    cp data/lang_phone/words.txt $lang_dir
+    if [ ! -f $lang_dir/transcript_words.txt ]; then
+      log "Generate text for BPE training from data/fbank/cuts_train.jsonl.gz"
+      python local/prepare_transcripts.py --cut ${fbank}/cuts_train.jsonl.gz --langdir ${lang_dir} 
+    fi
+    ./local/train_bpe_model.py \
+      --lang-dir $lang_dir \
+      --vocab-size $vocab_size \
+      --transcript $lang_dir/transcript_words.txt
+  done
+fi
+
+
+if [ $stage -le 7 ] && [ $stop_stage -ge 7 ]; then
+  log "Stage 7: Prepare BPE ST based lang"
+  for vocab_size in ${st_vocab_sizes[@]}; do
+    lang_dir=data/lang_st_bpe_${vocab_size}
+    mkdir -p ${lang_dir}
+
+    if [ ! -f $lang_dir/st_words.txt ]; then
+      log "Generate text for BPE training from data/fbank/cuts_train.jsonl.gz"
+      python local/prepare_st_transcripts.py --cut ${fbank}/cuts_train.jsonl.gz --langdir ${lang_dir} 
+    fi
+    ./local/train_bpe_model.py \
+      --lang-dir $lang_dir \
+      --vocab-size $vocab_size \
+      --transcript $lang_dir/st_words.txt
+  done
+fi
+

egs/multi_conv_zh_es_ta/ST/shared

@@ -0,0 +1 @@
+../../../icefall/shared

egs/multi_conv_zh_es_ta/ST/zipformer_multijoiner_st/asr_datamodule.py

@@ -0,0 +1,421 @@
+# Copyright      2021  Piotr Żelasko
+# Copyright      2022  Xiaomi Corporation     (Author: Mingshuang Luo)
+# Copyright      2025  Johns Hopkins University (Author: Amir Hussein)
+# See ../../../../LICENSE for clarification regarding multiple authors
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+
+import argparse
+import inspect
+import logging
+from functools import lru_cache
+from pathlib import Path
+from typing import Any, Dict, Optional
+
+import torch
+from lhotse import CutSet, Fbank, FbankConfig, load_manifest, load_manifest_lazy
+from lhotse.dataset import (  # noqa F401 for PrecomputedFeatures
+    CutConcatenate,
+    CutMix,
+    DynamicBucketingSampler,
+    K2SpeechRecognitionDataset,
+    K2Speech2TextTranslationDataset,
+    PrecomputedFeatures,
+    SimpleCutSampler,
+    SpecAugment,
+)
+from lhotse.dataset.input_strategies import (  # noqa F401 For AudioSamples
+    AudioSamples,
+    OnTheFlyFeatures,
+)
+from lhotse.utils import fix_random_seed
+from torch.utils.data import DataLoader
+
+from icefall.utils import str2bool
+
+
+class _SeedWorkers:
+    def __init__(self, seed: int):
+        self.seed = seed
+
+    def __call__(self, worker_id: int):
+        fix_random_seed(self.seed + worker_id)
+
+
+class MultiLingAsrDataModule:
+    """
+    DataModule for k2 ASR experiments.
+    It assumes there is always one train and valid dataloader,
+    but there can be multiple test dataloaders (e.g. LibriSpeech test-clean
+    and test-other).
+
+    It contains all the common data pipeline modules used in ASR
+    experiments, e.g.:
+    - dynamic batch size,
+    - bucketing samplers,
+    - cut concatenation,
+    - augmentation,
+    - on-the-fly feature extraction
+
+    This class should be derived for specific corpora used in ASR tasks.
+    """
+
+    def __init__(self, args: argparse.Namespace):
+        self.args = args
+
+    @classmethod
+    def add_arguments(cls, parser: argparse.ArgumentParser):
+        group = parser.add_argument_group(
+            title="ASR data related options",
+            description="These options are used for the preparation of "
+            "PyTorch DataLoaders from Lhotse CutSet's -- they control the "
+            "effective batch sizes, sampling strategies, applied data "
+            "augmentations, etc.",
+        )
+
+        group.add_argument(
+            "--manifest-dir",
+            type=Path,
+            default=Path("data/fbank"),
+            help="Path to directory with train/valid/test cuts.",
+        )
+        group.add_argument(
+            "--max-duration",
+            type=int,
+            default=200.0,
+            help="Maximum pooled recordings duration (seconds) in a "
+            "single batch. You can reduce it if it causes CUDA OOM.",
+        )
+        group.add_argument(
+            "--bucketing-sampler",
+            type=str2bool,
+            default=True,
+            help="When enabled, the batches will come from buckets of "
+            "similar duration (saves padding frames).",
+        )
+        group.add_argument(
+            "--num-buckets",
+            type=int,
+            default=30,
+            help="The number of buckets for the DynamicBucketingSampler"
+            "(you might want to increase it for larger datasets).",
+        )
+        group.add_argument(
+            "--concatenate-cuts",
+            type=str2bool,
+            default=False,
+            help="When enabled, utterances (cuts) will be concatenated "
+            "to minimize the amount of padding.",
+        )
+        group.add_argument(
+            "--duration-factor",
+            type=float,
+            default=1.0,
+            help="Determines the maximum duration of a concatenated cut "
+            "relative to the duration of the longest cut in a batch.",
+        )
+        group.add_argument(
+            "--gap",
+            type=float,
+            default=1.0,
+            help="The amount of padding (in seconds) inserted between "
+            "concatenated cuts. This padding is filled with noise when "
+            "noise augmentation is used.",
+        )
+        group.add_argument(
+            "--on-the-fly-feats",
+            type=str2bool,
+            default=False,
+            help="When enabled, use on-the-fly cut mixing and feature "
+            "extraction. Will drop existing precomputed feature manifests "
+            "if available.",
+        )
+        group.add_argument(
+            "--shuffle",
+            type=str2bool,
+            default=True,
+            help="When enabled (=default), the examples will be "
+            "shuffled for each epoch.",
+        )
+        group.add_argument(
+            "--drop-last",
+            type=str2bool,
+            default=True,
+            help="Whether to drop last batch. Used by sampler.",
+        )
+        group.add_argument(
+            "--return-cuts",
+            type=str2bool,
+            default=True,
+            help="When enabled, each batch will have the "
+            "field: batch['supervisions']['cut'] with the cuts that "
+            "were used to construct it.",
+        )
+
+        group.add_argument(
+            "--num-workers",
+            type=int,
+            default=10,
+            help="The number of training dataloader workers that "
+            "collect the batches.",
+        )
+
+        group.add_argument(
+            "--enable-spec-aug",
+            type=str2bool,
+            default=True,
+            help="When enabled, use SpecAugment for training dataset.",
+        )
+
+        group.add_argument(
+            "--spec-aug-time-warp-factor",
+            type=int,
+            default=80,
+            help="Used only when --enable-spec-aug is True. "
+            "It specifies the factor for time warping in SpecAugment. "
+            "Larger values mean more warping. "
+            "A value less than 1 means to disable time warp.",
+        )
+
+        group.add_argument(
+            "--enable-musan",
+            type=str2bool,
+            default=True,
+            help="When enabled, select noise from MUSAN and mix it"
+            "with training dataset. ",
+        )
+
+        group.add_argument(
+            "--input-strategy",
+            type=str,
+            default="PrecomputedFeatures",
+            help="AudioSamples or PrecomputedFeatures",
+        )
+
+    def train_dataloaders(
+        self,
+        cuts_train: CutSet,
+        sampler_state_dict: Optional[Dict[str, Any]] = None,
+    ) -> DataLoader:
+        """
+        Args:
+          cuts_train:
+            CutSet for training.
+          sampler_state_dict:
+            The state dict for the training sampler.
+        """
+        transforms = []
+        if self.args.enable_musan:
+            logging.info("Enable MUSAN")
+            logging.info("About to get Musan cuts")
+            cuts_musan = load_manifest(self.args.manifest_dir / "musan_cuts.jsonl.gz")
+            transforms.append(
+                CutMix(cuts=cuts_musan, p=0.5, snr=(10, 20), preserve_id=True)
+            )
+        else:
+            logging.info("Disable MUSAN")
+
+        if self.args.concatenate_cuts:
+            logging.info(
+                f"Using cut concatenation with duration factor "
+                f"{self.args.duration_factor} and gap {self.args.gap}."
+            )
+            # Cut concatenation should be the first transform in the list,
+            # so that if we e.g. mix noise in, it will fill the gaps between
+            # different utterances.
+            transforms = [
+                CutConcatenate(
+                    duration_factor=self.args.duration_factor, gap=self.args.gap
+                )
+            ] + transforms
+
+        input_transforms = []
+        if self.args.enable_spec_aug:
+            logging.info("Enable SpecAugment")
+            logging.info(f"Time warp factor: {self.args.spec_aug_time_warp_factor}")
+            # Set the value of num_frame_masks according to Lhotse's version.
+            # In different Lhotse's versions, the default of num_frame_masks is
+            # different.
+            num_frame_masks = 10
+            num_frame_masks_parameter = inspect.signature(
+                SpecAugment.__init__
+            ).parameters["num_frame_masks"]
+            if num_frame_masks_parameter.default == 1:
+                num_frame_masks = 2
+            logging.info(f"Num frame mask: {num_frame_masks}")
+            input_transforms.append(
+                SpecAugment(
+                    time_warp_factor=self.args.spec_aug_time_warp_factor,
+                    num_frame_masks=num_frame_masks,
+                    features_mask_size=27,
+                    num_feature_masks=2,
+                    frames_mask_size=100,
+                )
+            )
+        else:
+            logging.info("Disable SpecAugment")
+
+        logging.info("About to create train dataset")
+        train = K2Speech2TextTranslationDataset(
+            input_strategy=eval(self.args.input_strategy)(),
+            cut_transforms=transforms,
+            input_transforms=input_transforms,
+            return_cuts=self.args.return_cuts,
+        )
+
+        if self.args.on_the_fly_feats:
+            # NOTE: the PerturbSpeed transform should be added only if we
+            # remove it from data prep stage.
+            # Add on-the-fly speed perturbation; since originally it would
+            # have increased epoch size by 3, we will apply prob 2/3 and use
+            # 3x more epochs.
+            # Speed perturbation probably should come first before
+            # concatenation, but in principle the transforms order doesn't have
+            # to be strict (e.g. could be randomized)
+            # transforms = [PerturbSpeed(factors=[0.9, 1.1], p=2/3)] + transforms   # noqa
+            # Drop feats to be on the safe side.
+            train = K2Speech2TextTranslationDataset(
+                cut_transforms=transforms,
+                input_strategy=OnTheFlyFeatures(Fbank(FbankConfig(num_mel_bins=80))),
+                input_transforms=input_transforms,
+                return_cuts=self.args.return_cuts,
+            )
+
+        if self.args.bucketing_sampler:
+            logging.info("Using DynamicBucketingSampler.")
+            train_sampler = DynamicBucketingSampler(
+                cuts_train,
+                max_duration=self.args.max_duration,
+                shuffle=self.args.shuffle,
+                num_buckets=self.args.num_buckets,
+                drop_last=self.args.drop_last,
+            )
+        else:
+            logging.info("Using SimpleCutSampler.")
+            train_sampler = SimpleCutSampler(
+                cuts_train,
+                max_duration=self.args.max_duration,
+                shuffle=self.args.shuffle,
+            )
+        logging.info("About to create train dataloader")
+
+        if sampler_state_dict is not None:
+            logging.info("Loading sampler state dict")
+            train_sampler.load_state_dict(sampler_state_dict)
+
+        # 'seed' is derived from the current random state, which will have
+        # previously been set in the main process.
+        seed = torch.randint(0, 100000, ()).item()
+        worker_init_fn = _SeedWorkers(seed)
+
+        train_dl = DataLoader(
+            train,
+            sampler=train_sampler,
+            batch_size=None,
+            num_workers=self.args.num_workers,
+            persistent_workers=False,
+            worker_init_fn=worker_init_fn,
+        )
+
+        return train_dl
+
+    def valid_dataloaders(self, cuts_valid: CutSet) -> DataLoader:
+        transforms = []
+        if self.args.concatenate_cuts:
+            transforms = [
+                CutConcatenate(
+                    duration_factor=self.args.duration_factor, gap=self.args.gap
+                )
+            ] + transforms
+
+        logging.info("About to create dev dataset")
+        if self.args.on_the_fly_feats:
+            validate = K2Speech2TextTranslationDataset(
+                cut_transforms=transforms,
+                input_strategy=OnTheFlyFeatures(Fbank(FbankConfig(num_mel_bins=80))),
+                return_cuts=self.args.return_cuts,
+            )
+        else:
+            validate = K2Speech2TextTranslationDataset(
+                cut_transforms=transforms,
+                return_cuts=self.args.return_cuts,
+            )
+        valid_sampler = DynamicBucketingSampler(
+            cuts_valid,
+            max_duration=self.args.max_duration,
+            shuffle=False,
+        )
+        logging.info("About to create dev dataloader")
+        valid_dl = DataLoader(
+            validate,
+            sampler=valid_sampler,
+            batch_size=None,
+            num_workers=self.args.num_workers,
+            persistent_workers=False,
+        )
+
+        return valid_dl
+
+    def test_dataloaders(self, cuts: CutSet) -> DataLoader:
+        logging.debug("About to create test dataset")
+        test = K2Speech2TextTranslationDataset(
+            input_strategy=OnTheFlyFeatures(Fbank(FbankConfig(num_mel_bins=80)))
+            if self.args.on_the_fly_feats
+            else eval(self.args.input_strategy)(),
+            return_cuts=self.args.return_cuts,
+        )
+        sampler = DynamicBucketingSampler(
+            cuts,
+            max_duration=self.args.max_duration,
+            shuffle=False,
+        )
+        logging.debug("About to create test dataloader")
+        test_dl = DataLoader(
+            test,
+            batch_size=None,
+            sampler=sampler,
+            num_workers=self.args.num_workers,
+        )
+        return test_dl
+
+    @lru_cache()
+    def train_cuts(self) -> CutSet:
+        logging.info("Train data: About to get training cuts")
+        return load_manifest_lazy(
+            self.args.manifest_dir / "cuts_train.jsonl.gz"
+        )
+
+    @lru_cache()
+    def dev_all_cuts(self) -> CutSet:
+        logging.info("Dev data: About to get develop cuts")
+        return load_manifest_lazy(
+            self.args.manifest_dir / "cuts_dev.jsonl.gz"
+        )
+
+    @lru_cache()
+    def test_hkust(self) -> CutSet:
+        logging.info("About to get test-hkust cuts")
+        return load_manifest_lazy(
+            self.args.manifest_dir / "cuts_hkust_test.jsonl.gz"
+        )
+    def test_iwslt22(self) -> CutSet:
+        logging.info("About to get test-iwslt22 cuts")
+        return load_manifest_lazy(
+            self.args.manifest_dir / "cuts_iwslt_ta_test.jsonl.gz"
+        )
+    def test_fisher(self) -> CutSet:
+        logging.info("About to get test-fisher cuts")
+        return load_manifest_lazy(
+            self.args.manifest_dir / "cuts_fisher-sp_test.jsonl.gz"
+        ) 

egs/multi_conv_zh_es_ta/ST/zipformer_multijoiner_st/decoder.py

@@ -0,0 +1,131 @@
+# Copyright      2025  Johns Hopkins University (author: Amir Hussein)
+# Copyright    2021  Xiaomi Corp.        (authors: Fangjun Kuang)
+#
+# See ../../../../LICENSE for clarification regarding multiple authors
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+from scaling import Balancer
+
+
+class Decoder(nn.Module):
+    """This class modifies the stateless decoder from the following paper:
+
+        RNN-transducer with stateless prediction network
+        https://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=9054419
+
+    It removes the recurrent connection from the decoder, i.e., the prediction
+    network. Different from the above paper, it adds an extra Conv1d
+    right after the embedding layer.
+
+    TODO: Implement https://arxiv.org/pdf/2109.07513.pdf
+    """
+
+    def __init__(
+        self,
+        vocab_size: int,
+        decoder_dim: int,
+        blank_id: int,
+        context_size: int,
+    ):
+        """
+        Args:
+          vocab_size:
+            Number of tokens of the modeling unit including blank.
+          decoder_dim:
+            Dimension of the input embedding, and of the decoder output.
+          blank_id:
+            The ID of the blank symbol.
+          context_size:
+            Number of previous words to use to predict the next word.
+            1 means bigram; 2 means trigram. n means (n+1)-gram.
+        """
+        super().__init__()
+
+        self.embedding = nn.Embedding(
+            num_embeddings=vocab_size,
+            embedding_dim=decoder_dim,
+        )
+        # the balancers are to avoid any drift in the magnitude of the
+        # embeddings, which would interact badly with parameter averaging.
+        self.balancer = Balancer(
+            decoder_dim,
+            channel_dim=-1,
+            min_positive=0.0,
+            max_positive=1.0,
+            min_abs=0.5,
+            max_abs=1.0,
+            prob=0.05,
+        )
+
+        self.blank_id = blank_id
+
+        assert context_size >= 1, context_size
+        self.context_size = context_size
+        self.vocab_size = vocab_size
+
+        if context_size > 1:
+            self.conv = nn.Conv1d(
+                in_channels=decoder_dim,
+                out_channels=decoder_dim,
+                kernel_size=context_size,
+                padding=0,
+                groups=decoder_dim // 4,  # group size == 4
+                bias=False,
+            )
+            self.balancer2 = Balancer(
+                decoder_dim,
+                channel_dim=-1,
+                min_positive=0.0,
+                max_positive=1.0,
+                min_abs=0.5,
+                max_abs=1.0,
+                prob=0.05,
+            )
+
+    def forward(self, y: torch.Tensor, need_pad: bool = True) -> torch.Tensor:
+        """
+        Args:
+          y:
+            A 2-D tensor of shape (N, U).
+          need_pad:
+            True to left pad the input. Should be True during training.
+            False to not pad the input. Should be False during inference.
+        Returns:
+          Return a tensor of shape (N, U, decoder_dim).
+        """
+        y = y.to(torch.int64)
+        # this stuff about clamp() is a temporary fix for a mismatch
+        # at utterance start, we use negative ids in beam_search.py
+        embedding_out = self.embedding(y.clamp(min=0)) * (y >= 0).unsqueeze(-1)
+
+        embedding_out = self.balancer(embedding_out)
+
+        if self.context_size > 1:
+            embedding_out = embedding_out.permute(0, 2, 1)
+            if need_pad is True:
+                embedding_out = F.pad(embedding_out, pad=(self.context_size - 1, 0))
+            else:
+                # During inference time, there is no need to do extra padding
+                # as we only need one output
+                assert embedding_out.size(-1) == self.context_size
+            embedding_out = self.conv(embedding_out)
+            embedding_out = embedding_out.permute(0, 2, 1)
+            embedding_out = F.relu(embedding_out)
+            embedding_out = self.balancer2(embedding_out)
+
+        return embedding_out

egs/multi_conv_zh_es_ta/ST/zipformer_multijoiner_st/export.py

@@ -0,0 +1,590 @@
+#!/usr/bin/env python3
+# Copyright      2025  Johns Hopkins University (author: Amir Hussein)
+# Copyright 2021-2023 Xiaomi Corporation (Author: Fangjun Kuang,
+#                                                 Zengwei Yao,
+#                                                 Wei Kang)
+#
+# See ../../../../LICENSE for clarification regarding multiple authors
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+# This script converts several saved checkpoints
+# to a single one using model averaging.
+"""
+
+Usage:
+
+Note: This is a example for librispeech dataset, if you are using different
+dataset, you should change the argument values according to your dataset.
+
+(1) Export to torchscript model using torch.jit.script()
+
+- For non-streaming model:
+
+./zipformer/export.py \
+    --exp-dir ./zipformer_hat_st/exp-multi-joiner-nohat-pbe4k \
+    --bpe-st-model data/lang_st_bpe_4000/bpe.model \
+    --bpe-model data/lang_bpe_5000/bpe.model \
+    --epoch 25 \
+    --avg 13 \
+    --jit 1 \
+    --num-encoder-layers 2,2,2,2,2,2,2,2 \
+    --feedforward-dim  512,768,1024,1024,1024,1024,1024,768 \
+    --encoder-dim  192,256,384,512,384,384,384,256 \
+    --encoder-unmasked-dim 192,192,256,256,256,256,256,192 \
+    --downsampling-factor 1,2,4,8,8,4,4,2 \
+    --cnn-module-kernel 31,31,15,15,15,15,31,31 \
+    --num-heads 4,4,4,8,8,8,4,4 \
+    --use-hat False
+
+It will generate a file `jit_script.pt` in the given `exp_dir`. You can later
+load it by `torch.jit.load("jit_script.pt")`.
+
+Check ./jit_pretrained.py for its usage.
+
+Check https://github.com/k2-fsa/sherpa
+for how to use the exported models outside of icefall.
+
+- For streaming model:
+
+./zipformer_multijoiner_st/export.py \
+    --exp-dir ./zipformer_multijoiner_st/exp-multi-joiner-nohat-pbe4k_causal \
+    --bpe-st-model data/lang_st_bpe_4000/bpe.model \
+    --bpe-model data/lang_bpe_5000/bpe.model \
+    --causal 1 \
+    --chunk-size 64 \
+    --left-context-frames 128 \
+    --epoch 25 \
+    --avg 13 \
+    --jit 1 \
+    --num-encoder-layers 2,2,2,2,2,2,2,2 \
+    --feedforward-dim  512,768,1024,1024,1024,1024,1024,768 \
+    --encoder-dim  192,256,384,512,384,384,384,256 \
+    --encoder-unmasked-dim 192,192,256,256,256,256,256,192 \
+    --downsampling-factor 1,2,4,8,8,4,4,2 \
+    --cnn-module-kernel 31,31,15,15,15,15,31,31 \
+    --num-heads 4,4,4,8,8,8,4,4 \
+    --use-hat False
+
+It will generate a file `jit_script_chunk_16_left_128.pt` in the given `exp_dir`.
+You can later load it by `torch.jit.load("jit_script_chunk_16_left_128.pt")`.
+
+Check ./jit_pretrained_streaming.py for its usage.
+
+Check https://github.com/k2-fsa/sherpa
+for how to use the exported models outside of icefall.
+
+(2) Export `model.state_dict()`
+
+- For non-streaming model:
+
+./zipformer_multijoiner_st/export.py \
+    --exp-dir ./zipformer_multijoiner_st/exp-multi-joiner-nohat-pbe4k \
+    --bpe-st-model data/lang_st_bpe_4000/bpe.model \
+    --bpe-model data/lang_bpe_5000/bpe.model \
+    --epoch 25 \
+    --avg 13 \
+    --beam-size 20 \
+    --num-encoder-layers 2,2,2,2,2,2,2,2 \
+    --feedforward-dim  512,768,1024,1024,1024,1024,1024,768 \
+    --encoder-dim  192,256,384,512,384,384,384,256 \
+    --encoder-unmasked-dim 192,192,256,256,256,256,256,192 \
+    --downsampling-factor 1,2,4,8,8,4,4,2 \
+    --cnn-module-kernel 31,31,15,15,15,15,31,31 \
+    --num-heads 4,4,4,8,8,8,4,4 
+
+- For streaming model:
+
+./zipformer_multijoiner_st/export.py \
+    --exp-dir ./zipformer_multijoiner_st/exp-multi-joiner-nohat-pbe4k_causal \
+    --bpe-st-model data/lang_st_bpe_4000/bpe.model \
+    --bpe-model data/lang_bpe_5000/bpe.model \
+    --epoch 25 \
+    --avg 13 \
+    --causal 1 \
+    --chunk-size 64 \
+    --left-context-frames 128 \
+    --num-encoder-layers 2,2,2,2,2,2,2,2 \
+    --feedforward-dim  512,768,1024,1024,1024,1024,1024,768 \
+    --encoder-dim  192,256,384,512,384,384,384,256 \
+    --encoder-unmasked-dim 192,192,256,256,256,256,256,192 \
+    --downsampling-factor 1,2,4,8,8,4,4,2 \
+    --cnn-module-kernel 31,31,15,15,15,15,31,31 \
+    --num-heads 4,4,4,8,8,8,4,4 
+
+It will generate a file `pretrained.pt` in the given `exp_dir`. You can later
+load it by `icefall.checkpoint.load_checkpoint()`.
+
+- For non-streaming model:
+
+To use the generated file with `zipformer/decode.py`,
+you can do:
+
+    cd /path/to/exp_dir
+    ln -s pretrained.pt epoch-9999.pt
+
+    cd /path/to/egs/librispeech/ASR
+    ./zipformer_multijoiner_st/decode.py \
+        --exp-dir ./zipformer_multijoiner_st/exp-multi-joiner-nohat-pbe4k \
+        --epoch 9999 \
+        --avg 1 \
+        --beam-size 20 \
+        --max-duration 600 \
+        --decoding-method modified_beam_search \
+        --bpe-st-model data/lang_st_bpe_4000/bpe.model \
+        --bpe-model data/lang_bpe_5000/bpe.model \
+        --num-encoder-layers 2,2,2,2,2,2,2,2 \
+        --feedforward-dim  512,768,1024,1024,1024,1024,1024,768 \
+        --encoder-dim  192,256,384,512,384,384,384,256 \
+        --encoder-unmasked-dim 192,192,256,256,256,256,256,192 \
+        --downsampling-factor 1,2,4,8,8,4,4,2 \
+        --cnn-module-kernel 31,31,15,15,15,15,31,31 \
+        --num-heads 4,4,4,8,8,8,4,4 \
+        --use-averaged-model false
+
+- For streaming model:
+
+To use the generated file with `zipformer/decode.py` and `zipformer/streaming_decode.py`, you can do:
+
+    cd /path/to/exp_dir
+    ln -s pretrained.pt epoch-9999.pt
+
+    cd /path/to/egs/librispeech/ASR
+
+    # simulated streaming decoding
+    ./zipformer_multijoiner_st/decode.py \
+        --exp-dir ./zipformer_multijoiner_st/exp-multi-joiner-nohat-pbe4k \
+        --epoch 9999 \
+        --avg 1 \
+        --causal 1 \
+        --beam-size 20 \
+        --max-duration 600 \
+        --bpe-st-model data/lang_st_bpe_4000/bpe.model \
+        --bpe-model data/lang_bpe_5000/bpe.model \
+        --num-encoder-layers 2,2,2,2,2,2,2,2 \
+        --feedforward-dim  512,768,1024,1024,1024,1024,1024,768 \
+        --encoder-dim  192,256,384,512,384,384,384,256 \
+        --encoder-unmasked-dim 192,192,256,256,256,256,256,192 \
+        --downsampling-factor 1,2,4,8,8,4,4,2 \
+        --cnn-module-kernel 31,31,15,15,15,15,31,31 \
+        --num-heads 4,4,4,8,8,8,4,4 \
+        --use-averaged-model false \
+        --chunk-size 64 \
+        --left-context-frames 128 \
+        --max-sym-per-frame 20  \
+        --use-hat False \
+        --decoding-method greedy_search 
+
+Check ./pretrained.py for its usage.
+
+Note: If you don't want to train a model from scratch, we have
+provided one for you. You can get it at
+
+- non-streaming model:
+https://huggingface.co/Zengwei/icefall-asr-librispeech-zipformer-2023-05-15
+
+- streaming model:
+https://huggingface.co/Zengwei/icefall-asr-librispeech-streaming-zipformer-2023-05-17
+
+with the following commands:
+
+    sudo apt-get install git-lfs
+    git lfs install
+    git clone https://huggingface.co/Zengwei/icefall-asr-librispeech-zipformer-2023-05-15
+    git clone https://huggingface.co/Zengwei/icefall-asr-librispeech-streaming-zipformer-2023-05-17
+    # You will find the pre-trained models in exp dir
+"""
+
+import argparse
+import logging
+from pathlib import Path
+from typing import List, Tuple
+
+import k2
+import sentencepiece as spm
+import torch
+from scaling_converter import convert_scaled_to_non_scaled
+from torch import Tensor, nn
+from train import add_model_arguments, get_model, get_params
+
+from icefall.checkpoint import (
+    average_checkpoints,
+    average_checkpoints_with_averaged_model,
+    find_checkpoints,
+    load_checkpoint,
+)
+from icefall.utils import make_pad_mask, num_tokens, str2bool
+
+
+def get_parser():
+    parser = argparse.ArgumentParser(
+        formatter_class=argparse.ArgumentDefaultsHelpFormatter
+    )
+
+    parser.add_argument(
+        "--epoch",
+        type=int,
+        default=30,
+        help="""It specifies the checkpoint to use for decoding.
+        Note: Epoch counts from 1.
+        You can specify --avg to use more checkpoints for model averaging.""",
+    )
+
+    parser.add_argument(
+        "--iter",
+        type=int,
+        default=0,
+        help="""If positive, --epoch is ignored and it
+        will use the checkpoint exp_dir/checkpoint-iter.pt.
+        You can specify --avg to use more checkpoints for model averaging.
+        """,
+    )
+
+    parser.add_argument(
+        "--avg",
+        type=int,
+        default=9,
+        help="Number of checkpoints to average. Automatically select "
+        "consecutive checkpoints before the checkpoint specified by "
+        "'--epoch' and '--iter'",
+    )
+
+    parser.add_argument(
+        "--use-averaged-model",
+        type=str2bool,
+        default=True,
+        help="Whether to load averaged model. Currently it only supports "
+        "using --epoch. If True, it would decode with the averaged model "
+        "over the epoch range from `epoch-avg` (excluded) to `epoch`."
+        "Actually only the models with epoch number of `epoch-avg` and "
+        "`epoch` are loaded for averaging. ",
+    )
+
+    parser.add_argument(
+        "--exp-dir",
+        type=str,
+        default="zipformer/exp",
+        help="""It specifies the directory where all training related
+        files, e.g., checkpoints, log, etc, are saved
+        """,
+    )
+
+    parser.add_argument(
+        "--bpe-st-model",
+        type=str,
+        default="data/lang_st_bpe_4000/bpe.model",
+        help="Path to the BPE model",
+    )
+    parser.add_argument(
+        "--bpe-model",
+        type=str,
+        default="data/lang_bpe_5000/bpe.model",
+        help="Path to the BPE model",
+    )
+
+    parser.add_argument(
+        "--jit",
+        type=str2bool,
+        default=False,
+        help="""True to save a model after applying torch.jit.script.
+        It will generate a file named jit_script.pt.
+        Check ./jit_pretrained.py for how to use it.
+        """,
+    )
+
+    parser.add_argument(
+        "--context-size",
+        type=int,
+        default=2,
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
+    )
+
+    add_model_arguments(parser)
+
+    return parser
+
+
+class EncoderModel(nn.Module):
+    """A wrapper for encoder and encoder_embed"""
+
+    def __init__(self, encoder: nn.Module, encoder_embed: nn.Module) -> None:
+        super().__init__()
+        self.encoder = encoder
+        self.encoder_embed = encoder_embed
+
+    def forward(
+        self, features: Tensor, feature_lengths: Tensor
+    ) -> Tuple[Tensor, Tensor]:
+        """
+        Args:
+            features: (N, T, C)
+            feature_lengths: (N,)
+        """
+        x, x_lens = self.encoder_embed(features, feature_lengths)
+
+        src_key_padding_mask = make_pad_mask(x_lens)
+        x = x.permute(1, 0, 2)  # (N, T, C) -> (T, N, C)
+
+        encoder_out, encoder_out_lens = self.encoder(x, x_lens, src_key_padding_mask)
+        encoder_out = encoder_out.permute(1, 0, 2)  # (T, N, C) ->(N, T, C)
+
+        return encoder_out, encoder_out_lens
+
+
+class StreamingEncoderModel(nn.Module):
+    """A wrapper for encoder and encoder_embed"""
+
+    def __init__(self, encoder: nn.Module, encoder_embed: nn.Module) -> None:
+        super().__init__()
+        assert len(encoder.chunk_size) == 1, encoder.chunk_size
+        assert len(encoder.left_context_frames) == 1, encoder.left_context_frames
+        self.chunk_size = encoder.chunk_size[0]
+        self.left_context_len = encoder.left_context_frames[0]
+
+        # The encoder_embed subsample features (T - 7) // 2
+        # The ConvNeXt module needs (7 - 1) // 2 = 3 frames of right padding after subsampling
+        self.pad_length = 7 + 2 * 3
+
+        self.encoder = encoder
+        self.encoder_embed = encoder_embed
+
+    def forward(
+        self, features: Tensor, feature_lengths: Tensor, states: List[Tensor]
+    ) -> Tuple[Tensor, Tensor, List[Tensor]]:
+        """Streaming forward for encoder_embed and encoder.
+
+        Args:
+            features: (N, T, C)
+            feature_lengths: (N,)
+            states: a list of Tensors
+
+        Returns encoder outputs, output lengths, and updated states.
+        """
+        chunk_size = self.chunk_size
+        left_context_len = self.left_context_len
+
+        cached_embed_left_pad = states[-2]
+        x, x_lens, new_cached_embed_left_pad = self.encoder_embed.streaming_forward(
+            x=features,
+            x_lens=feature_lengths,
+            cached_left_pad=cached_embed_left_pad,
+        )
+        assert x.size(1) == chunk_size, (x.size(1), chunk_size)
+
+        src_key_padding_mask = make_pad_mask(x_lens)
+
+        # processed_mask is used to mask out initial states
+        processed_mask = torch.arange(left_context_len, device=x.device).expand(
+            x.size(0), left_context_len
+        )
+        processed_lens = states[-1]  # (batch,)
+        # (batch, left_context_size)
+        processed_mask = (processed_lens.unsqueeze(1) <= processed_mask).flip(1)
+        # Update processed lengths
+        new_processed_lens = processed_lens + x_lens
+
+        # (batch, left_context_size + chunk_size)
+        src_key_padding_mask = torch.cat([processed_mask, src_key_padding_mask], dim=1)
+
+        x = x.permute(1, 0, 2)  # (N, T, C) -> (T, N, C)
+        encoder_states = states[:-2]
+
+        (
+            encoder_out,
+            encoder_out_lens,
+            new_encoder_states,
+        ) = self.encoder.streaming_forward(
+            x=x,
+            x_lens=x_lens,
+            states=encoder_states,
+            src_key_padding_mask=src_key_padding_mask,
+        )
+        encoder_out = encoder_out.permute(1, 0, 2)  # (T, N, C) ->(N, T, C)
+
+        new_states = new_encoder_states + [
+            new_cached_embed_left_pad,
+            new_processed_lens,
+        ]
+        return encoder_out, encoder_out_lens, new_states
+
+    @torch.jit.export
+    def get_init_states(
+        self,
+        batch_size: int = 1,
+        device: torch.device = torch.device("cpu"),
+    ) -> List[torch.Tensor]:
+        """
+        Returns a list of cached tensors of all encoder layers. For layer-i, states[i*6:(i+1)*6]
+        is (cached_key, cached_nonlin_attn, cached_val1, cached_val2, cached_conv1, cached_conv2).
+        states[-2] is the cached left padding for ConvNeXt module,
+        of shape (batch_size, num_channels, left_pad, num_freqs)
+        states[-1] is processed_lens of shape (batch,), which records the number
+        of processed frames (at 50hz frame rate, after encoder_embed) for each sample in batch.
+        """
+        states = self.encoder.get_init_states(batch_size, device)
+
+        embed_states = self.encoder_embed.get_init_states(batch_size, device)
+        states.append(embed_states)
+
+        processed_lens = torch.zeros(batch_size, dtype=torch.int32, device=device)
+        states.append(processed_lens)
+
+        return states
+
+
+@torch.no_grad()
+def main():
+    args = get_parser().parse_args()
+    args.exp_dir = Path(args.exp_dir)
+
+    params = get_params()
+    params.update(vars(args))
+
+    device = torch.device("cpu")
+    # if torch.cuda.is_available():
+    #     device = torch.device("cuda", 0)
+
+    logging.info(f"device: {device}")
+
+    sp = spm.SentencePieceProcessor()
+    sp.load(params.bpe_model)
+
+    sp_st = spm.SentencePieceProcessor()
+    sp_st.load(params.bpe_st_model)
+
+    # <blk> and <unk> are defined in local/train_bpe_model.py
+    params.blank_id = sp.piece_to_id("<blk>")
+    params.unk_id = sp.piece_to_id("<unk>")
+    params.blank_st_id = sp_st.piece_to_id("<blk>")
+    params.st_unk_id = sp_st.piece_to_id("<unk>")
+    params.vocab_size = sp.get_piece_size()
+    params.st_vocab_size = sp_st.get_piece_size()
+
+    logging.info(params)
+
+    logging.info("About to create model")
+    model = get_model(params)
+
+    if not params.use_averaged_model:
+        if params.iter > 0:
+            filenames = find_checkpoints(params.exp_dir, iteration=-params.iter)[
+                : params.avg
+            ]
+            if len(filenames) == 0:
+                raise ValueError(
+                    f"No checkpoints found for"
+                    f" --iter {params.iter}, --avg {params.avg}"
+                )
+            elif len(filenames) < params.avg:
+                raise ValueError(
+                    f"Not enough checkpoints ({len(filenames)}) found for"
+                    f" --iter {params.iter}, --avg {params.avg}"
+                )
+            logging.info(f"averaging {filenames}")
+            model.load_state_dict(average_checkpoints(filenames, device=device))
+        elif params.avg == 1:
+            load_checkpoint(f"{params.exp_dir}/epoch-{params.epoch}.pt", model)
+        else:
+            start = params.epoch - params.avg + 1
+            filenames = []
+            for i in range(start, params.epoch + 1):
+                if i >= 1:
+                    filenames.append(f"{params.exp_dir}/epoch-{i}.pt")
+            logging.info(f"averaging {filenames}")
+            model.load_state_dict(average_checkpoints(filenames, device=device))
+    else:
+        if params.iter > 0:
+            filenames = find_checkpoints(params.exp_dir, iteration=-params.iter)[
+                : params.avg + 1
+            ]
+            if len(filenames) == 0:
+                raise ValueError(
+                    f"No checkpoints found for"
+                    f" --iter {params.iter}, --avg {params.avg}"
+                )
+            elif len(filenames) < params.avg + 1:
+                raise ValueError(
+                    f"Not enough checkpoints ({len(filenames)}) found for"
+                    f" --iter {params.iter}, --avg {params.avg}"
+                )
+            filename_start = filenames[-1]
+            filename_end = filenames[0]
+            logging.info(
+                "Calculating the averaged model over iteration checkpoints"
+                f" from {filename_start} (excluded) to {filename_end}"
+            )
+            model.load_state_dict(
+                average_checkpoints_with_averaged_model(
+                    filename_start=filename_start,
+                    filename_end=filename_end,
+                    device=device,
+                )
+            )
+        elif params.avg == 1:
+            load_checkpoint(f"{params.exp_dir}/epoch-{params.epoch}.pt", model)
+        else:
+            assert params.avg > 0, params.avg
+            start = params.epoch - params.avg
+            assert start >= 1, start
+            filename_start = f"{params.exp_dir}/epoch-{start}.pt"
+            filename_end = f"{params.exp_dir}/epoch-{params.epoch}.pt"
+            logging.info(
+                f"Calculating the averaged model over epoch range from "
+                f"{start} (excluded) to {params.epoch}"
+            )
+            model.load_state_dict(
+                average_checkpoints_with_averaged_model(
+                    filename_start=filename_start,
+                    filename_end=filename_end,
+                    device=device,
+                )
+            )
+
+    model.eval()
+
+    if params.jit is True:
+        convert_scaled_to_non_scaled(model, inplace=True)
+        # We won't use the forward() method of the model in C++, so just ignore
+        # it here.
+        # Otherwise, one of its arguments is a ragged tensor and is not
+        # torch scriptabe.
+        model.__class__.forward = torch.jit.ignore(model.__class__.forward)
+
+        # Wrap encoder and encoder_embed as a module
+        if params.causal:
+            model.encoder = StreamingEncoderModel(model.encoder, model.encoder_embed)
+            chunk_size = model.encoder.chunk_size
+            left_context_len = model.encoder.left_context_len
+            filename = f"jit_script_chunk_{chunk_size}_left_{left_context_len}.pt"
+        else:
+            model.encoder = EncoderModel(model.encoder, model.encoder_embed)
+            filename = "jit_script.pt"
+
+        logging.info("Using torch.jit.script")
+        model = torch.jit.script(model)
+        model.save(str(params.exp_dir / filename))
+        logging.info(f"Saved to {filename}")
+    else:
+        logging.info("Not using torchscript. Export model.state_dict()")
+        # Save it using a format so that it can be loaded
+        # by :func:`load_checkpoint`
+        filename = params.exp_dir / "pretrained.pt"
+        torch.save({"model": model.state_dict()}, str(filename))
+        logging.info(f"Saved to {filename}")
+
+
+if __name__ == "__main__":
+    formatter = "%(asctime)s %(levelname)s [%(filename)s:%(lineno)d] %(message)s"
+
+    logging.basicConfig(format=formatter, level=logging.INFO)
+    main()

egs/multi_conv_zh_es_ta/ST/zipformer_multijoiner_st/joiner.py

@@ -0,0 +1,72 @@
+# Copyright    2021  Xiaomi Corp.        (authors: Fangjun Kuang)
+#
+# See ../../../../LICENSE for clarification regarding multiple authors
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import torch
+import torch.nn as nn
+from scaling import ScaledLinear
+from typing import Optional
+
+class Joiner(nn.Module):
+    def __init__(
+        self,
+        encoder_dim: int,
+        decoder_dim: int,
+        joiner_dim: int,
+        vocab_size: int,
+        encoder_lid: Optional[int] = None,
+    ):
+        super().__init__()
+        self.encoder_proj = ScaledLinear(encoder_dim, joiner_dim, initial_scale=0.25)
+        self.decoder_proj = ScaledLinear(decoder_dim, joiner_dim, initial_scale=0.25)
+        if encoder_lid:
+            self.lid_proj = ScaledLinear(encoder_lid, joiner_dim, initial_scale=0.25)
+        self.output_linear = nn.Linear(joiner_dim, vocab_size) 
+
+    def forward(
+        self,
+        encoder_out: torch.Tensor,
+        decoder_out: torch.Tensor,
+        project_input: bool = True,
+        lid_out: Optional[torch.Tensor] = None,
+    ) -> torch.Tensor:
+        """
+        Args:
+          encoder_out:
+            Output from the encoder. Its shape is (N, T, s_range, C).
+          decoder_out:
+            Output from the decoder. Its shape is (N, T, s_range, C).
+           project_input:
+            If true, apply input projections encoder_proj and decoder_proj.
+            If this is false, it is the user's responsibility to do this
+            manually.
+        Returns:
+          Return a tensor of shape (N, T, s_range, C).
+        """
+        assert encoder_out.ndim == decoder_out.ndim, (
+            encoder_out.shape,
+            decoder_out.shape,
+        )
+
+        if project_input:
+            logit = self.encoder_proj(encoder_out) + self.decoder_proj(decoder_out)
+        elif lid_out is not None:
+            logit = encoder_out + decoder_out + lid_out
+        else:
+            logit = encoder_out + decoder_out
+
+        logit = self.output_linear(torch.tanh(logit))
+
+        return logit

egs/multi_conv_zh_es_ta/ST/zipformer_multijoiner_st/model.py

@@ -0,0 +1,681 @@
+# Copyright      2025  Johns Hopkins University (author: Amir Hussein)
+# Copyright    2021-2023  Xiaomi Corp.        (authors: Fangjun Kuang,
+#                                                       Wei Kang,
+#                                                       Zengwei Yao)
+#
+# See ../../../../LICENSE for clarification regarding multiple authors
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from typing import Optional, Tuple, Union
+
+import k2
+import torch
+from torch import Tensor
+from lhotse.dataset import SpecAugment
+import torch.nn as nn
+from encoder_interface import EncoderInterface
+
+from icefall.utils import  add_sos, make_pad_mask, time_warp
+from scaling import ScaledLinear
+
+
+class StModel(nn.Module):
+    def __init__(
+        self,
+        encoder_embed: nn.Module,
+        encoder: EncoderInterface,
+        decoder: Optional[nn.Module] = None,
+        joiner: Optional[nn.Module] = None,
+        st_joiner: Optional[nn.Module] = None,
+        st_decoder: Optional[nn.Module] = None,
+        encoder_dim: int = 384,
+        decoder_dim: int = 512,
+        vocab_size: int = 500,
+        st_vocab_size: int = 500,
+        use_transducer: bool = True,
+        use_ctc: bool = False,
+        use_st_ctc: bool = False,
+        use_hat: bool = False,
+    ):
+        """A multitask Transducer ASR-ST model with seperate joiners and predictors but shared acoustic encoder.
+
+        - Connectionist temporal classification: labelling unsegmented sequence data with recurrent neural networks (http://imagine.enpc.fr/~obozinsg/teaching/mva_gm/papers/ctc.pdf)
+        - Sequence Transduction with Recurrent Neural Networks (https://arxiv.org/pdf/1211.3711.pdf)
+        - Pruned RNN-T for fast, memory-efficient ASR training (https://arxiv.org/pdf/2206.13236.pdf)
+
+        Args:
+          encoder_embed:
+            It is a Convolutional 2D subsampling module. It converts
+            an input of shape (N, T, idim) to an output of of shape
+            (N, T', odim), where T' = (T-3)//2-2 = (T-7)//2.
+          encoder:
+            It is the transcription network in the paper. Its accepts
+            two inputs: `x` of (N, T, encoder_dim) and `x_lens` of shape (N,).
+            It returns two tensors: `logits` of shape (N, T, encoder_dim) and
+            `logit_lens` of shape (N,).
+          decoder:
+            It is the prediction network in the paper. Its input shape
+            is (N, U) and its output shape is (N, U, decoder_dim).
+            It should contain one attribute: `blank_id`.
+            It is used when use_transducer is True.
+          joiner:
+            It has two inputs with shapes: (N, T, encoder_dim) and (N, U, decoder_dim).
+            Its output shape is (N, T, U, vocab_size). Note that its output contains
+            unnormalized probs, i.e., not processed by log-softmax.
+            It is used when use_transducer is True.
+          use_transducer:
+            Whether use transducer head. Default: True.
+          use_ctc:
+            Whether use CTC head. Default: False.
+        """
+        super().__init__()
+
+        assert (
+            use_transducer or use_ctc
+        ), f"At least one of them should be True, but got use_transducer={use_transducer}, use_ctc={use_ctc}"
+
+        assert isinstance(encoder, EncoderInterface), type(encoder)
+
+        self.encoder_embed = encoder_embed
+        self.encoder = encoder
+        self.use_hat = use_hat
+        self.use_transducer = use_transducer
+        if use_transducer:
+            # Modules for Transducer head
+            assert decoder is not None
+            assert hasattr(decoder, "blank_id")
+            assert joiner is not None
+
+            self.decoder = decoder
+            self.joiner = joiner
+            
+            self.st_joiner = st_joiner
+            self.st_decoder = st_decoder
+
+            self.simple_am_proj = ScaledLinear(
+                encoder_dim, vocab_size, initial_scale=0.25
+            )
+            self.simple_lm_proj = ScaledLinear(
+                decoder_dim, vocab_size, initial_scale=0.25
+            )
+
+            self.simple_st_am_proj = ScaledLinear(
+                encoder_dim, st_vocab_size, initial_scale=0.25
+            )
+            self.simple_st_lm_proj = ScaledLinear(
+                decoder_dim, st_vocab_size, initial_scale=0.25
+            )
+        else:
+            assert decoder is None
+            assert joiner is None
+
+        self.use_ctc = use_ctc
+        self.use_st_ctc = use_st_ctc
+        if self.use_ctc:
+            # Modules for CTC head
+            self.ctc_output = nn.Sequential(
+                nn.Dropout(p=0.1),
+                nn.Linear(encoder_dim, vocab_size),
+                nn.LogSoftmax(dim=-1),
+            )
+        if self.use_st_ctc:
+            # Modules for CTC head
+            self.st_ctc_output = nn.Sequential(
+                nn.Dropout(p=0.1),
+                nn.Linear(encoder_dim, st_vocab_size),
+                nn.LogSoftmax(dim=-1),
+            )
+
+    def forward_encoder(
+        self, x: torch.Tensor, x_lens: torch.Tensor
+    ) -> Tuple[torch.Tensor, torch.Tensor]:
+        """Compute encoder outputs.
+        Args:
+          x:
+            A 3-D tensor of shape (N, T, C).
+          x_lens:
+            A 1-D tensor of shape (N,). It contains the number of frames in `x`
+            before padding.
+
+        Returns:
+          encoder_out:
+            Encoder output, of shape (N, T, C).
+          encoder_out_lens:
+            Encoder output lengths, of shape (N,).
+        """
+        # logging.info(f"Memory allocated at entry: {torch.cuda.memory_allocated() // 1000000}M")
+        x, x_lens = self.encoder_embed(x, x_lens)
+        # logging.info(f"Memory allocated after encoder_embed: {torch.cuda.memory_allocated() // 1000000}M")
+
+        src_key_padding_mask = make_pad_mask(x_lens)
+        x = x.permute(1, 0, 2)  # (N, T, C) -> (T, N, C)
+
+        encoder_out, encoder_out_lens = self.encoder(x, x_lens, src_key_padding_mask)
+
+        encoder_out = encoder_out.permute(1, 0, 2)  # (T, N, C) ->(N, T, C)
+        assert torch.all(encoder_out_lens > 0), (x_lens, encoder_out_lens)
+        return encoder_out, encoder_out_lens
+    def forward_st_ctc(
+        self,
+        st_encoder_out: torch.Tensor,
+        st_encoder_out_lens: torch.Tensor,
+        targets: torch.Tensor,
+        target_lengths: torch.Tensor,
+    ) -> torch.Tensor:
+        """Compute CTC loss.
+        Args:
+          encoder_out:
+            Encoder output, of shape (N, T, C).
+          encoder_out_lens:
+            Encoder output lengths, of shape (N,).
+          targets:
+            Target Tensor of shape (sum(target_lengths)). The targets are assumed
+            to be un-padded and concatenated within 1 dimension.
+        """
+        # Compute CTC log-prob
+        ctc_output = self.st_ctc_output(st_encoder_out)  # (N, T, C)
+
+        ctc_loss = torch.nn.functional.ctc_loss(
+            log_probs=ctc_output.permute(1, 0, 2),  # (T, N, C)
+            targets=targets,
+            input_lengths=st_encoder_out_lens,
+            target_lengths=target_lengths,
+            reduction="sum",
+        )
+        return ctc_loss
+
+    def forward_ctc(
+        self,
+        encoder_out: torch.Tensor,
+        encoder_out_lens: torch.Tensor,
+        targets: torch.Tensor,
+        target_lengths: torch.Tensor,
+    ) -> torch.Tensor:
+        """Compute CTC loss.
+        Args:
+          encoder_out:
+            Encoder output, of shape (N, T, C).
+          encoder_out_lens:
+            Encoder output lengths, of shape (N,).
+          targets:
+            Target Tensor of shape (sum(target_lengths)). The targets are assumed
+            to be un-padded and concatenated within 1 dimension.
+        """
+        # Compute CTC log-prob
+        ctc_output = self.ctc_output(encoder_out)  # (N, T, C)
+
+        ctc_loss = torch.nn.functional.ctc_loss(
+            log_probs=ctc_output.permute(1, 0, 2),  # (T, N, C)
+            targets=targets,
+            input_lengths=encoder_out_lens,
+            target_lengths=target_lengths,
+            reduction="sum",
+        )
+        return ctc_loss
+      
+    def forward_cr_ctc(
+        self,
+        encoder_out: torch.Tensor,
+        encoder_out_lens: torch.Tensor,
+        targets: torch.Tensor,
+        target_lengths: torch.Tensor,
+    ) -> Tuple[torch.Tensor, torch.Tensor]:
+        """Compute CTC loss with consistency regularization loss.
+        Args:
+          encoder_out:
+            Encoder output, of shape (2 * N, T, C).
+          encoder_out_lens:
+            Encoder output lengths, of shape (2 * N,).
+          targets:
+            Target Tensor of shape (2 * sum(target_lengths)). The targets are assumed
+            to be un-padded and concatenated within 1 dimension.
+        """
+        # Compute CTC loss
+        ctc_output = self.ctc_output(encoder_out)  # (2 * N, T, C)
+        ctc_loss = torch.nn.functional.ctc_loss(
+            log_probs=ctc_output.permute(1, 0, 2),  # (T, 2 * N, C)
+            targets=targets.cpu(),
+            input_lengths=encoder_out_lens.cpu(),
+            target_lengths=target_lengths.cpu(),
+            reduction="none",
+        )
+        ctc_loss_is_finite = torch.isfinite(ctc_loss)
+        ctc_loss = ctc_loss[ctc_loss_is_finite]
+        ctc_loss = ctc_loss.sum()
+
+        # Compute consistency regularization loss
+        exchanged_targets = ctc_output.detach().chunk(2, dim=0)
+        exchanged_targets = torch.cat(
+            [exchanged_targets[1], exchanged_targets[0]], dim=0
+        )  # exchange: [x1, x2] -> [x2, x1]
+        cr_loss = nn.functional.kl_div(
+            input=ctc_output,
+            target=exchanged_targets,
+            reduction="none",
+            log_target=True,
+        )  # (2 * N, T, C)
+        length_mask = make_pad_mask(encoder_out_lens).unsqueeze(-1)
+        cr_loss = cr_loss.masked_fill(length_mask, 0.0).sum()
+
+        return ctc_loss, cr_loss
+
+    def forward_st_cr_ctc(
+        self,
+        st_encoder_out: torch.Tensor,
+        st_encoder_out_lens: torch.Tensor,
+        st_targets: torch.Tensor,
+        st_target_lengths: torch.Tensor,
+        # encoder_out: torch.Tensor,
+        # encoder_out_lens: torch.Tensor,
+        # targets: torch.Tensor,
+        # target_lengths: torch.Tensor,
+    ) -> Tuple[torch.Tensor, torch.Tensor]:
+        """Compute CTC loss with consistency regularization loss.
+        Args:
+          encoder_out:
+            Encoder output, of shape (2 * N, T, C).
+          encoder_out_lens:
+            Encoder output lengths, of shape (2 * N,).
+          targets:
+            Target Tensor of shape (2 * sum(target_lengths)). The targets are assumed
+            to be un-padded and concatenated within 1 dimension.
+        """
+        # Compute CTC loss
+        st_ctc_output = self.st_ctc_output(st_encoder_out)  # (2 * N, T, C)
+        st_ctc_loss = torch.nn.functional.ctc_loss(
+            log_probs=st_ctc_output.permute(1, 0, 2),  # (T, 2 * N, C)
+            targets=st_targets.cpu(),
+            input_lengths=st_encoder_out_lens.cpu(),
+            target_lengths=st_target_lengths.cpu(),
+            reduction="none",
+        )
+        st_ctc_loss_is_finite = torch.isfinite(st_ctc_loss)
+        st_ctc_loss = st_ctc_loss[st_ctc_loss_is_finite]
+        st_ctc_loss = st_ctc_loss.sum()
+        # ctc_output = self.ctc_output(encoder_out)  # (2 * N, T, C)
+        # ctc_loss = torch.nn.functional.ctc_loss(
+        #     log_probs=ctc_output.permute(1, 0, 2),  # (T, 2 * N, C)
+        #     targets=targets.cpu(),
+        #     input_lengths=encoder_out_lens.cpu(),
+        #     target_lengths=target_lengths.cpu(),
+        #     reduction="sum",
+        # )
+        
+        # if not torch.isfinite(st_ctc_loss):
+        #             breakpoint()
+
+        # Compute consistency regularization loss
+        exchanged_targets = st_ctc_output.detach().chunk(2, dim=0)
+        exchanged_targets = torch.cat(
+            [exchanged_targets[1], exchanged_targets[0]], dim=0
+        )  # exchange: [x1, x2] -> [x2, x1]
+        cr_loss = nn.functional.kl_div(
+            input=st_ctc_output,
+            target=exchanged_targets,
+            reduction="none",
+            log_target=True,
+        )  # (2 * N, T, C)
+        length_mask = make_pad_mask(st_encoder_out_lens).unsqueeze(-1)
+        cr_loss = cr_loss.masked_fill(length_mask, 0.0).sum()
+
+        return st_ctc_loss, cr_loss
+
+    def forward_transducer(
+        self,
+        encoder_out: torch.Tensor,
+        encoder_out_lens: torch.Tensor,
+        y: k2.RaggedTensor,
+        y_lens: torch.Tensor,
+        st_y: k2.RaggedTensor,
+        st_y_lens: torch.Tensor,
+        prune_range: int = 5,
+        am_scale: float = 0.0,
+        lm_scale: float = 0.0,
+    ) -> Union[Tuple[Tensor, Tensor], Tuple[Tensor, Tensor, Tensor]]:
+        """Compute Transducer loss.
+        Args:
+          encoder_out:
+            Encoder output, of shape (N, T, C).
+          encoder_out_lens:
+            Encoder output lengths, of shape (N,).
+          y:
+            A ragged tensor with 2 axes [utt][label]. It contains labels of each
+            utterance.
+          prune_range:
+            The prune range for rnnt loss, it means how many symbols(context)
+            we are considering for each frame to compute the loss.
+          am_scale:
+            The scale to smooth the loss with am (output of encoder network)
+            part
+          lm_scale:
+            The scale to smooth the loss with lm (output of predictor network)
+            part
+        """
+        # Now for the decoder, i.e., the prediction network
+        
+
+        blank_id = self.decoder.blank_id
+        st_blank_id = self.st_decoder.blank_id
+        sos_y = add_sos(y, sos_id=blank_id)
+        st_sos_y = add_sos(st_y, sos_id=st_blank_id)
+
+        # sos_y_padded: [B, S + 1], start with SOS.
+        sos_y_padded = sos_y.pad(mode="constant", padding_value=blank_id)
+        st_sos_y_padded = st_sos_y.pad(mode="constant", padding_value=st_blank_id)
+
+        # decoder_out: [B, S + 1, decoder_dim]
+        decoder_out = self.decoder(sos_y_padded)
+        st_decoder_out = self.st_decoder(st_sos_y_padded)
+
+        # Note: y does not start with SOS
+        # y_padded : [B, S]
+        y_padded = y.pad(mode="constant", padding_value=0)
+
+        y_padded = y_padded.to(torch.int64)
+        boundary = torch.zeros(
+            (encoder_out.size(0), 4),
+            dtype=torch.int64,
+            device=encoder_out.device,
+        )
+        boundary[:, 2] = y_lens
+        boundary[:, 3] = encoder_out_lens
+
+        st_y_padded = st_y.pad(mode="constant", padding_value=0)
+
+        st_y_padded = st_y_padded.to(torch.int64)
+        st_boundary = torch.zeros(
+            (encoder_out.size(0), 4),
+            dtype=torch.int64,
+            device=encoder_out.device,
+        )
+        st_boundary[:, 2] = st_y_lens
+        st_boundary[:, 3] = encoder_out_lens
+
+        lm = self.simple_lm_proj(decoder_out)
+        am = self.simple_am_proj(encoder_out)
+        st_lm = self.simple_st_lm_proj(st_decoder_out)
+        st_am = self.simple_st_am_proj(encoder_out)
+
+        # if self.training and random.random() < 0.25:
+        #    lm = penalize_abs_values_gt(lm, 100.0, 1.0e-04)
+        # if self.training and random.random() < 0.25:
+        #    am = penalize_abs_values_gt(am, 30.0, 1.0e-04)
+
+        with torch.cuda.amp.autocast(enabled=False):
+          simple_loss, (px_grad, py_grad) = k2.rnnt_loss_smoothed(
+              lm=lm.float(),
+              am=am.float(),
+              symbols=y_padded,
+              termination_symbol=blank_id,
+              lm_only_scale=lm_scale,
+              am_only_scale=am_scale,
+              boundary=boundary,
+              reduction="sum",
+              return_grad=True,
+          )
+          st_simple_loss, (st_px_grad, st_py_grad) = k2.rnnt_loss_smoothed(
+              lm=st_lm.float(),
+              am=st_am.float(),
+              symbols=st_y_padded,
+              termination_symbol=st_blank_id,
+              lm_only_scale=lm_scale,
+              am_only_scale=am_scale,
+              boundary=st_boundary,
+              reduction="sum",
+              return_grad=True,
+          )
+          
+
+        # am_pruned : [B, T, prune_range, encoder_dim]
+        # lm_pruned : [B, T, prune_range, decoder_dim]
+       
+        # ranges : [B, T, prune_range]
+        ranges = k2.get_rnnt_prune_ranges(
+            px_grad=px_grad,
+            py_grad=py_grad,
+            boundary=boundary,
+            s_range=prune_range,
+        )
+        am_pruned, lm_pruned = k2.do_rnnt_pruning(
+            am=self.joiner.encoder_proj(encoder_out),
+            lm=self.joiner.decoder_proj(decoder_out),
+            ranges=ranges,
+        )
+          # project_input=False since we applied the decoder's input projections
+          # prior to do_rnnt_pruning (this is an optimization for speed).
+        logits = self.joiner(am_pruned, lm_pruned, project_input=False)
+        with torch.cuda.amp.autocast(enabled=False):
+          pruned_loss = k2.rnnt_loss_pruned(
+              logits=logits.float(),
+              symbols=y_padded,
+              ranges=ranges,
+              termination_symbol=blank_id,
+              boundary=boundary,
+              reduction="sum",
+              use_hat_loss=self.use_hat,
+          )
+         # logits : [B, T, prune_range, vocab_size]
+        
+        st_ranges = k2.get_rnnt_prune_ranges(
+          px_grad=st_px_grad,
+          py_grad=st_py_grad,
+          boundary=st_boundary,
+          s_range=prune_range,
+        )
+        st_am_pruned, st_lm_pruned = k2.do_rnnt_pruning(
+            am=self.st_joiner.encoder_proj(encoder_out),
+            lm=self.st_joiner.decoder_proj(st_decoder_out),
+            ranges=st_ranges,
+        )
+
+        st_logits = self.st_joiner(st_am_pruned, st_lm_pruned, project_input=False)
+        # Compute HAT loss for st
+        with torch.cuda.amp.autocast(enabled=False):
+          pruned_st_loss = k2.rnnt_loss_pruned(
+              logits=st_logits.float(),
+              symbols=st_y.pad(mode="constant", padding_value=blank_id).to(torch.int64),
+              ranges=st_ranges,
+              termination_symbol=st_blank_id,
+              boundary=st_boundary,
+              reduction="sum",
+              use_hat_loss=self.use_hat,
+          )
+        
+        return simple_loss, st_simple_loss, pruned_loss, pruned_st_loss
+
+    def forward(
+        self,
+        x: torch.Tensor,
+        x_lens: torch.Tensor,
+        y: k2.RaggedTensor,
+        st_y: k2.RaggedTensor,
+        prune_range: int = 5,
+        am_scale: float = 0.0,
+        lm_scale: float = 0.0,
+        use_st_cr_ctc: bool = False,
+        use_asr_cr_ctc: bool = False,
+        use_spec_aug: bool = False,
+        spec_augment: Optional[SpecAugment] = None,
+        supervision_segments: Optional[torch.Tensor] = None,
+        time_warp_factor: Optional[int] = 80,
+    ) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor]:
+        """
+        Args:
+          x:
+            A 3-D tensor of shape (N, T, C).
+          x_lens:
+            A 1-D tensor of shape (N,). It contains the number of frames in `x`
+            before padding.
+          y:
+            A ragged tensor with 2 axes [utt][label]. It contains labels of each
+            utterance.
+          prune_range:
+            The prune range for rnnt loss, it means how many symbols(context)
+            we are considering for each frame to compute the loss.
+          am_scale:
+            The scale to smooth the loss with am (output of encoder network)
+            part
+          lm_scale:
+            The scale to smooth the loss with lm (output of predictor network)
+            part
+          use_cr_ctc:
+            Whether use consistency-regularized CTC.
+          use_spec_aug:
+            Whether apply spec-augment manually, used only if use_cr_ctc is True.
+          spec_augment:
+            The SpecAugment instance that returns time masks,
+            used only if use_cr_ctc is True.
+          supervision_segments:
+            An int tensor of shape ``(S, 3)``. ``S`` is the number of
+            supervision segments that exist in ``features``.
+            Used only if use_cr_ctc is True.
+          time_warp_factor:
+            Parameter for the time warping; larger values mean more warping.
+            Set to ``None``, or less than ``1``, to disable.
+            Used only if use_cr_ctc is True.
+        Returns:
+          Return the transducer losses and CTC loss,
+          in form of (simple_loss, pruned_loss, ctc_loss)
+
+        Note:
+           Regarding am_scale & lm_scale, it will make the loss-function one of
+           the form:
+              lm_scale * lm_probs + am_scale * am_probs +
+              (1-lm_scale-am_scale) * combined_probs
+        """
+        assert x.ndim == 3, x.shape
+        assert x_lens.ndim == 1, x_lens.shape
+        assert y.num_axes == 2, y.num_axes
+        assert st_y.num_axes == 2, st_y.num_axes
+
+        assert x.size(0) == x_lens.size(0) == y.dim0, (x.shape, x_lens.shape, y.dim0)
+
+        if use_st_cr_ctc or use_asr_cr_ctc:
+          assert self.use_ctc or self.use_st_ctc
+          if use_spec_aug:
+              assert spec_augment is not None and spec_augment.time_warp_factor < 1
+              # Apply time warping before input duplicating
+              assert supervision_segments is not None
+              x = time_warp(
+                  x,
+                  time_warp_factor=time_warp_factor,
+                  supervision_segments=supervision_segments,
+              )
+              # Independently apply frequency masking and time masking to the two copies
+              x = spec_augment(x.repeat(2, 1, 1))
+          else:
+              x = x.repeat(2, 1, 1)
+          x_lens = x_lens.repeat(2)
+          y = k2.ragged.cat([y, y], axis=0)
+          if self.st_joiner != None and self.use_st_ctc:
+            st_y = k2.ragged.cat([st_y, st_y], axis=0)
+
+        # Compute encoder outputs
+
+        encoder_out, encoder_out_lens = self.forward_encoder(x, x_lens)
+        row_splits = y.shape.row_splits(1)
+        y_lens = row_splits[1:] - row_splits[:-1]
+        st_row_splits = st_y.shape.row_splits(1)
+        st_y_lens = st_row_splits[1:] - st_row_splits[:-1]
+
+        if self.use_transducer:
+            
+            # Compute transducer loss
+            if self.st_joiner != None:
+              simple_loss, st_simple_loss, pruned_loss, st_pruned_loss = self.forward_transducer(
+                  encoder_out=encoder_out,
+                  encoder_out_lens=encoder_out_lens,
+                  y=y.to(x.device),
+                  y_lens=y_lens,
+                  st_y=st_y.to(x.device),
+                  st_y_lens=st_y_lens,
+                  prune_range=prune_range,
+                  am_scale=am_scale,
+                  lm_scale=lm_scale,
+              )
+              if use_asr_cr_ctc:
+                simple_loss = simple_loss * 0.5
+                pruned_loss = pruned_loss * 0.5
+              if use_st_cr_ctc:
+                st_simple_loss = st_simple_loss * 0.5
+                st_pruned_loss = st_pruned_loss * 0.5
+            else:
+                simple_loss, pruned_loss = self.forward_transducer(
+                  encoder_out=encoder_out,
+                  encoder_out_lens=encoder_out_lens,
+                  y=y.to(x.device),
+                  y_lens=y_lens,
+                  prune_range=prune_range,
+                  am_scale=am_scale,
+                  lm_scale=lm_scale,
+              )
+                if use_asr_cr_ctc:
+                  simple_loss = simple_loss * 0.5
+                  pruned_loss = pruned_loss * 0.5
+                st_simple_loss, st_pruned_loss = torch.empty(0), torch.empty(0)
+        else:
+            simple_loss = torch.empty(0)
+            pruned_loss = torch.empty(0)
+
+        if self.use_ctc:
+            # Compute CTC loss
+            targets = y.values
+            if not use_asr_cr_ctc:
+                ctc_loss = self.forward_ctc(
+                    encoder_out=encoder_out,
+                    encoder_out_lens=encoder_out_lens,
+                    targets=targets,
+                    target_lengths=y_lens,
+                )
+                cr_loss = torch.empty(0)
+            else:
+                ctc_loss, cr_loss = self.forward_cr_ctc(
+                    encoder_out=encoder_out,
+                    encoder_out_lens=encoder_out_lens,
+                    targets=targets,
+                    target_lengths=y_lens,
+                )
+                ctc_loss = ctc_loss * 0.5
+                cr_loss = cr_loss * 0.5
+        else:
+            cr_loss = torch.empty(0)
+            ctc_loss = torch.empty(0)
+        if self.use_st_ctc:
+            st_targets = st_y.values
+            if not use_st_cr_ctc:
+              st_ctc_loss = self.forward_st_ctc(
+                    st_encoder_out=encoder_out,
+                    st_encoder_out_lens=encoder_out_lens,
+                    targets=st_targets,
+                    target_lengths=st_y_lens,
+                )
+              st_cr_loss = torch.empty(0)
+            else:
+              st_ctc_loss, st_cr_loss = self.forward_st_cr_ctc(
+                  st_encoder_out=encoder_out,
+                  st_encoder_out_lens=encoder_out_lens,
+                  st_targets=st_targets,
+                  st_target_lengths=st_y_lens,
+                  # encoder_out=encoder_out,
+                  # encoder_out_lens=encoder_out_lens,
+                  # targets=targets,
+                  # target_lengths=y_lens,
+              )
+              st_ctc_loss = st_ctc_loss * 0.5
+              st_cr_loss = st_cr_loss * 0.5
+        else:
+            st_ctc_loss = torch.empty(0)
+            st_cr_loss = torch.empty(0)
+
+        return simple_loss, st_simple_loss, pruned_loss, st_pruned_loss, ctc_loss, st_ctc_loss, cr_loss, st_cr_loss

egs/multi_conv_zh_es_ta/ST/zipformer_multijoiner_st/pretrained.py

@@ -0,0 +1,381 @@
+#!/usr/bin/env python3
+# Copyright      2021-2023  Xiaomi Corp.        (authors: Fangjun Kuang, Zengwei Yao)
+#
+# See ../../../../LICENSE for clarification regarding multiple authors
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This script loads a checkpoint and uses it to decode waves.
+You can generate the checkpoint with the following command:
+
+Note: This is a example for librispeech dataset, if you are using different
+dataset, you should change the argument values according to your dataset.
+
+- For non-streaming model:
+
+./zipformer/export.py \
+  --exp-dir ./zipformer/exp \
+  --tokens data/lang_bpe_500/tokens.txt \
+  --epoch 30 \
+  --avg 9
+
+- For streaming model:
+
+./zipformer/export.py \
+  --exp-dir ./zipformer/exp \
+  --causal 1 \
+  --tokens data/lang_bpe_500/tokens.txt \
+  --epoch 30 \
+  --avg 9
+
+Usage of this script:
+
+- For non-streaming model:
+
+(1) greedy search
+./zipformer/pretrained.py \
+  --checkpoint ./zipformer/exp/pretrained.pt \
+  --tokens data/lang_bpe_500/tokens.txt \
+  --method greedy_search \
+  /path/to/foo.wav \
+  /path/to/bar.wav
+
+(2) modified beam search
+./zipformer/pretrained.py \
+  --checkpoint ./zipformer/exp/pretrained.pt \
+  --tokens ./data/lang_bpe_500/tokens.txt \
+  --method modified_beam_search \
+  /path/to/foo.wav \
+  /path/to/bar.wav
+
+(3) fast beam search
+./zipformer/pretrained.py \
+  --checkpoint ./zipformer/exp/pretrained.pt \
+  --tokens ./data/lang_bpe_500/tokens.txt \
+  --method fast_beam_search \
+  /path/to/foo.wav \
+  /path/to/bar.wav
+
+- For streaming model:
+
+(1) greedy search
+./zipformer/pretrained.py \
+  --checkpoint ./zipformer/exp/pretrained.pt \
+  --causal 1 \
+  --chunk-size 16 \
+  --left-context-frames 128 \
+  --tokens ./data/lang_bpe_500/tokens.txt \
+  --method greedy_search \
+  /path/to/foo.wav \
+  /path/to/bar.wav
+
+(2) modified beam search
+./zipformer/pretrained.py \
+  --checkpoint ./zipformer/exp/pretrained.pt \
+  --causal 1 \
+  --chunk-size 16 \
+  --left-context-frames 128 \
+  --tokens ./data/lang_bpe_500/tokens.txt \
+  --method modified_beam_search \
+  /path/to/foo.wav \
+  /path/to/bar.wav
+
+(3) fast beam search
+./zipformer/pretrained.py \
+  --checkpoint ./zipformer/exp/pretrained.pt \
+  --causal 1 \
+  --chunk-size 16 \
+  --left-context-frames 128 \
+  --tokens ./data/lang_bpe_500/tokens.txt \
+  --method fast_beam_search \
+  /path/to/foo.wav \
+  /path/to/bar.wav
+
+
+You can also use `./zipformer/exp/epoch-xx.pt`.
+
+Note: ./zipformer/exp/pretrained.pt is generated by ./zipformer/export.py
+"""
+
+
+import argparse
+import logging
+import math
+from typing import List
+
+import k2
+import kaldifeat
+import torch
+import torchaudio
+from beam_search import (
+    fast_beam_search_one_best,
+    greedy_search_batch,
+    modified_beam_search,
+)
+from export import num_tokens
+from torch.nn.utils.rnn import pad_sequence
+from train import add_model_arguments, get_model, get_params
+
+from icefall.utils import make_pad_mask
+
+
+def get_parser():
+    parser = argparse.ArgumentParser(
+        formatter_class=argparse.ArgumentDefaultsHelpFormatter
+    )
+
+    parser.add_argument(
+        "--checkpoint",
+        type=str,
+        required=True,
+        help="Path to the checkpoint. "
+        "The checkpoint is assumed to be saved by "
+        "icefall.checkpoint.save_checkpoint().",
+    )
+
+    parser.add_argument(
+        "--tokens",
+        type=str,
+        help="""Path to tokens.txt.""",
+    )
+
+    parser.add_argument(
+        "--method",
+        type=str,
+        default="greedy_search",
+        help="""Possible values are:
+          - greedy_search
+          - modified_beam_search
+          - fast_beam_search
+        """,
+    )
+
+    parser.add_argument(
+        "sound_files",
+        type=str,
+        nargs="+",
+        help="The input sound file(s) to transcribe. "
+        "Supported formats are those supported by torchaudio.load(). "
+        "For example, wav and flac are supported. "
+        "The sample rate has to be 16kHz.",
+    )
+
+    parser.add_argument(
+        "--sample-rate",
+        type=int,
+        default=16000,
+        help="The sample rate of the input sound file",
+    )
+
+    parser.add_argument(
+        "--beam-size",
+        type=int,
+        default=4,
+        help="""An integer indicating how many candidates we will keep for each
+        frame. Used only when --method is beam_search or
+        modified_beam_search.""",
+    )
+
+    parser.add_argument(
+        "--beam",
+        type=float,
+        default=4,
+        help="""A floating point value to calculate the cutoff score during beam
+        search (i.e., `cutoff = max-score - beam`), which is the same as the
+        `beam` in Kaldi.
+        Used only when --method is fast_beam_search""",
+    )
+
+    parser.add_argument(
+        "--max-contexts",
+        type=int,
+        default=4,
+        help="""Used only when --method is fast_beam_search""",
+    )
+
+    parser.add_argument(
+        "--max-states",
+        type=int,
+        default=8,
+        help="""Used only when --method is fast_beam_search""",
+    )
+
+    parser.add_argument(
+        "--context-size",
+        type=int,
+        default=2,
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
+    )
+
+    parser.add_argument(
+        "--max-sym-per-frame",
+        type=int,
+        default=1,
+        help="""Maximum number of symbols per frame. Used only when
+        --method is greedy_search.
+        """,
+    )
+
+    add_model_arguments(parser)
+
+    return parser
+
+
+def read_sound_files(
+    filenames: List[str], expected_sample_rate: float
+) -> List[torch.Tensor]:
+    """Read a list of sound files into a list 1-D float32 torch tensors.
+    Args:
+      filenames:
+        A list of sound filenames.
+      expected_sample_rate:
+        The expected sample rate of the sound files.
+    Returns:
+      Return a list of 1-D float32 torch tensors.
+    """
+    ans = []
+    for f in filenames:
+        wave, sample_rate = torchaudio.load(f)
+        assert (
+            sample_rate == expected_sample_rate
+        ), f"expected sample rate: {expected_sample_rate}. Given: {sample_rate}"
+        # We use only the first channel
+        ans.append(wave[0].contiguous())
+    return ans
+
+
+@torch.no_grad()
+def main():
+    parser = get_parser()
+    args = parser.parse_args()
+
+    params = get_params()
+
+    params.update(vars(args))
+
+    token_table = k2.SymbolTable.from_file(params.tokens)
+
+    params.blank_id = token_table["<blk>"]
+    params.unk_id = token_table["<unk>"]
+    params.vocab_size = num_tokens(token_table) + 1
+
+    logging.info(f"{params}")
+
+    device = torch.device("cpu")
+    if torch.cuda.is_available():
+        device = torch.device("cuda", 0)
+
+    logging.info(f"device: {device}")
+
+    if params.causal:
+        assert (
+            "," not in params.chunk_size
+        ), "chunk_size should be one value in decoding."
+        assert (
+            "," not in params.left_context_frames
+        ), "left_context_frames should be one value in decoding."
+
+    logging.info("Creating model")
+    model = get_model(params)
+
+    num_param = sum([p.numel() for p in model.parameters()])
+    logging.info(f"Number of model parameters: {num_param}")
+
+    checkpoint = torch.load(args.checkpoint, map_location="cpu")
+    model.load_state_dict(checkpoint["model"], strict=False)
+    model.to(device)
+    model.eval()
+
+    logging.info("Constructing Fbank computer")
+    opts = kaldifeat.FbankOptions()
+    opts.device = device
+    opts.frame_opts.dither = 0
+    opts.frame_opts.snip_edges = False
+    opts.frame_opts.samp_freq = params.sample_rate
+    opts.mel_opts.num_bins = params.feature_dim
+
+    fbank = kaldifeat.Fbank(opts)
+
+    logging.info(f"Reading sound files: {params.sound_files}")
+    waves = read_sound_files(
+        filenames=params.sound_files, expected_sample_rate=params.sample_rate
+    )
+    waves = [w.to(device) for w in waves]
+
+    logging.info("Decoding started")
+    features = fbank(waves)
+    feature_lengths = [f.size(0) for f in features]
+
+    features = pad_sequence(features, batch_first=True, padding_value=math.log(1e-10))
+    feature_lengths = torch.tensor(feature_lengths, device=device)
+
+    # model forward
+    encoder_out, encoder_out_lens = model.forward_encoder(features, feature_lengths)
+
+    hyps = []
+    msg = f"Using {params.method}"
+    logging.info(msg)
+
+    def token_ids_to_words(token_ids: List[int]) -> str:
+        text = ""
+        for i in token_ids:
+            text += token_table[i]
+        return text.replace("▁", " ").strip()
+
+    if params.method == "fast_beam_search":
+        decoding_graph = k2.trivial_graph(params.vocab_size - 1, device=device)
+        hyp_tokens = fast_beam_search_one_best(
+            model=model,
+            decoding_graph=decoding_graph,
+            encoder_out=encoder_out,
+            encoder_out_lens=encoder_out_lens,
+            beam=params.beam,
+            max_contexts=params.max_contexts,
+            max_states=params.max_states,
+        )
+        for hyp in hyp_tokens:
+            hyps.append(token_ids_to_words(hyp))
+    elif params.method == "modified_beam_search":
+        hyp_tokens = modified_beam_search(
+            model=model,
+            encoder_out=encoder_out,
+            encoder_out_lens=encoder_out_lens,
+            beam=params.beam_size,
+        )
+
+        for hyp in hyp_tokens:
+            hyps.append(token_ids_to_words(hyp))
+    elif params.method == "greedy_search" and params.max_sym_per_frame == 1:
+        hyp_tokens = greedy_search_batch(
+            model=model,
+            encoder_out=encoder_out,
+            encoder_out_lens=encoder_out_lens,
+        )
+        for hyp in hyp_tokens:
+            hyps.append(token_ids_to_words(hyp))
+    else:
+        raise ValueError(f"Unsupported method: {params.method}")
+
+    s = "\n"
+    for filename, hyp in zip(params.sound_files, hyps):
+        s += f"{filename}:\n{hyp}\n\n"
+    logging.info(s)
+
+    logging.info("Decoding Done")
+
+
+if __name__ == "__main__":
+    formatter = "%(asctime)s %(levelname)s [%(filename)s:%(lineno)d] %(message)s"
+
+    logging.basicConfig(format=formatter, level=logging.INFO)
+    main()

egs/multi_conv_zh_es_ta/ST/zipformer_multijoiner_st/profile.py

@@ -0,0 +1,170 @@
+#!/usr/bin/env python3
+#
+# Copyright 2023 Xiaomi Corporation     (Author: Zengwei Yao)
+#
+# See ../../../../LICENSE for clarification regarding multiple authors
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""
+Usage: ./zipformer/profile.py
+"""
+
+import argparse
+import logging
+import sentencepiece as spm
+import torch
+
+from typing import Tuple
+from torch import Tensor, nn
+
+from icefall.utils import make_pad_mask
+from icefall.profiler import get_model_profile
+from scaling import BiasNorm
+from train import (
+    get_encoder_embed,
+    get_encoder_model,
+    get_joiner_model,
+    add_model_arguments,
+    get_params,
+)
+from zipformer import BypassModule
+
+
+def get_parser():
+    parser = argparse.ArgumentParser(
+        formatter_class=argparse.ArgumentDefaultsHelpFormatter
+    )
+
+    parser.add_argument(
+        "--bpe-model",
+        type=str,
+        default="data/lang_bpe_500/bpe.model",
+        help="Path to the BPE model",
+    )
+
+    add_model_arguments(parser)
+
+    return parser
+
+
+def _bias_norm_flops_compute(module, input, output):
+    assert len(input) == 1, len(input)
+    # estimate as layer_norm, see icefall/profiler.py
+    flops = input[0].numel() * 5
+    module.__flops__ += int(flops)
+
+
+def _swoosh_module_flops_compute(module, input, output):
+    # For SwooshL and SwooshR modules
+    assert len(input) == 1, len(input)
+    # estimate as swish/silu, see icefall/profiler.py
+    flops = input[0].numel()
+    module.__flops__ += int(flops)
+
+
+def _bypass_module_flops_compute(module, input, output):
+    # For Bypass module
+    assert len(input) == 2, len(input)
+    flops = input[0].numel() * 2
+    module.__flops__ += int(flops)
+
+
+MODULE_HOOK_MAPPING = {
+    BiasNorm: _bias_norm_flops_compute,
+    BypassModule: _bypass_module_flops_compute,
+}
+
+
+class Model(nn.Module):
+    """A Wrapper for encoder, encoder_embed, and encoder_proj"""
+
+    def __init__(
+        self,
+        encoder: nn.Module,
+        encoder_embed: nn.Module,
+        encoder_proj: nn.Module,
+    ) -> None:
+        super().__init__()
+        self.encoder = encoder
+        self.encoder_embed = encoder_embed
+        self.encoder_proj = encoder_proj
+
+    def forward(self, feature: Tensor, feature_lens: Tensor) -> Tuple[Tensor, Tensor]:
+        x, x_lens = self.encoder_embed(feature, feature_lens)
+
+        src_key_padding_mask = make_pad_mask(x_lens)
+        x = x.permute(1, 0, 2)  # (N, T, C) -> (T, N, C)
+
+        encoder_out, encoder_out_lens = self.encoder(x, x_lens, src_key_padding_mask)
+
+        encoder_out = encoder_out.permute(1, 0, 2)  # (N, T, C) -> (T, N, C)
+        logits = self.encoder_proj(encoder_out)
+
+        return logits, encoder_out_lens
+
+
+@torch.no_grad()
+def main():
+    parser = get_parser()
+    args = parser.parse_args()
+
+    params = get_params()
+    params.update(vars(args))
+
+    device = torch.device("cpu")
+    if torch.cuda.is_available():
+        device = torch.device("cuda", 0)
+    logging.info(f"Device: {device}")
+
+    sp = spm.SentencePieceProcessor()
+    sp.load(params.bpe_model)
+
+    # <blk> is defined in local/train_bpe_model.py
+    params.blank_id = sp.piece_to_id("<blk>")
+    params.vocab_size = sp.get_piece_size()
+
+    logging.info(params)
+
+    logging.info("About to create model")
+
+    # We only profile the encoder part
+    model = Model(
+        encoder=get_encoder_model(params),
+        encoder_embed=get_encoder_embed(params),
+        encoder_proj=get_joiner_model(params).encoder_proj,
+    )
+    model.eval()
+    model.to(device)
+
+    num_param = sum([p.numel() for p in model.parameters()])
+    logging.info(f"Number of model parameters: {num_param}")
+
+    # for 30-second input
+    B, T, D = 1, 3000, 80
+    feature = torch.ones(B, T, D, dtype=torch.float32).to(device)
+    feature_lens = torch.full((B,), T, dtype=torch.int64).to(device)
+
+    flops, params = get_model_profile(
+        model=model,
+        args=(feature, feature_lens),
+        module_hoop_mapping=MODULE_HOOK_MAPPING,
+    )
+    logging.info(f"For the encoder part, params: {params}, flops: {flops}")
+
+
+if __name__ == "__main__":
+    formatter = "%(asctime)s %(levelname)s [%(filename)s:%(lineno)d] %(message)s"
+    logging.basicConfig(format=formatter, level=logging.INFO)
+
+    main()

egs/multi_conv_zh_es_ta/ST/zipformer_multijoiner_st/scaling_converter.py

@@ -0,0 +1,104 @@
+# Copyright    2022-2023  Xiaomi Corp.        (authors: Fangjun Kuang, Zengwei Yao)
+#
+# See ../../../../LICENSE for clarification regarding multiple authors
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""
+This file replaces various modules in a model.
+Specifically, ActivationBalancer is replaced with an identity operator;
+Whiten is also replaced with an identity operator;
+BasicNorm is replaced by a module with `exp` removed.
+"""
+
+import copy
+from typing import List, Tuple
+
+import torch
+import torch.nn as nn
+from scaling import (
+    Balancer,
+    Dropout3,
+    ScaleGrad,
+    SwooshL,
+    SwooshLOnnx,
+    SwooshR,
+    SwooshROnnx,
+    Whiten,
+)
+from zipformer import CompactRelPositionalEncoding
+
+
+# Copied from https://pytorch.org/docs/1.9.0/_modules/torch/nn/modules/module.html#Module.get_submodule  # noqa
+# get_submodule was added to nn.Module at v1.9.0
+def get_submodule(model, target):
+    if target == "":
+        return model
+    atoms: List[str] = target.split(".")
+    mod: torch.nn.Module = model
+    for item in atoms:
+        if not hasattr(mod, item):
+            raise AttributeError(
+                mod._get_name() + " has no " "attribute `" + item + "`"
+            )
+        mod = getattr(mod, item)
+        if not isinstance(mod, torch.nn.Module):
+            raise AttributeError("`" + item + "` is not " "an nn.Module")
+    return mod
+
+
+def convert_scaled_to_non_scaled(
+    model: nn.Module,
+    inplace: bool = False,
+    is_pnnx: bool = False,
+    is_onnx: bool = False,
+):
+    """
+    Args:
+      model:
+        The model to be converted.
+      inplace:
+        If True, the input model is modified inplace.
+        If False, the input model is copied and we modify the copied version.
+      is_pnnx:
+        True if we are going to export the model for PNNX.
+      is_onnx:
+        True if we are going to export the model for ONNX.
+    Return:
+      Return a model without scaled layers.
+    """
+    if not inplace:
+        model = copy.deepcopy(model)
+
+    d = {}
+    for name, m in model.named_modules():
+        if isinstance(m, (Balancer, Dropout3, ScaleGrad, Whiten)):
+            d[name] = nn.Identity()
+        elif is_onnx and isinstance(m, SwooshR):
+            d[name] = SwooshROnnx()
+        elif is_onnx and isinstance(m, SwooshL):
+            d[name] = SwooshLOnnx()
+        elif is_onnx and isinstance(m, CompactRelPositionalEncoding):
+            # We want to recreate the positional encoding vector when
+            # the input changes, so we have to use torch.jit.script()
+            # to replace torch.jit.trace()
+            d[name] = torch.jit.script(m)
+
+    for k, v in d.items():
+        if "." in k:
+            parent, child = k.rsplit(".", maxsplit=1)
+            setattr(get_submodule(model, parent), child, v)
+        else:
+            setattr(model, k, v)
+
+    return model

egs/multi_conv_zh_es_ta/ST/zipformer_multijoiner_st/subsampling.py

@@ -0,0 +1,406 @@
+#!/usr/bin/env python3
+# Copyright    2023  Xiaomi Corp.        (authors: Daniel Povey,
+#                                                  Zengwei Yao)
+#
+# See ../../../../LICENSE for clarification regarding multiple authors
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from typing import Tuple
+import warnings
+
+import torch
+from torch import Tensor, nn
+from scaling import (
+    Balancer,
+    BiasNorm,
+    Dropout3,
+    FloatLike,
+    Optional,
+    ScaledConv2d,
+    ScaleGrad,
+    ScheduledFloat,
+    SwooshL,
+    SwooshR,
+    Whiten,
+)
+
+
+class ConvNeXt(nn.Module):
+    """
+    Our interpretation of the ConvNeXt module as used in https://arxiv.org/pdf/2206.14747.pdf
+    """
+
+    def __init__(
+        self,
+        channels: int,
+        hidden_ratio: int = 3,
+        kernel_size: Tuple[int, int] = (7, 7),
+        layerdrop_rate: FloatLike = None,
+    ):
+        super().__init__()
+        self.padding = ((kernel_size[0] - 1) // 2, (kernel_size[1] - 1) // 2)
+        hidden_channels = channels * hidden_ratio
+        if layerdrop_rate is None:
+            layerdrop_rate = ScheduledFloat((0.0, 0.2), (20000.0, 0.015))
+        self.layerdrop_rate = layerdrop_rate
+
+        self.depthwise_conv = nn.Conv2d(
+            in_channels=channels,
+            out_channels=channels,
+            groups=channels,
+            kernel_size=kernel_size,
+            padding=self.padding,
+        )
+
+        self.pointwise_conv1 = nn.Conv2d(
+            in_channels=channels, out_channels=hidden_channels, kernel_size=1
+        )
+
+        self.hidden_balancer = Balancer(
+            hidden_channels,
+            channel_dim=1,
+            min_positive=0.3,
+            max_positive=1.0,
+            min_abs=0.75,
+            max_abs=5.0,
+        )
+
+        self.activation = SwooshL()
+        self.pointwise_conv2 = ScaledConv2d(
+            in_channels=hidden_channels,
+            out_channels=channels,
+            kernel_size=1,
+            initial_scale=0.01,
+        )
+
+        self.out_balancer = Balancer(
+            channels,
+            channel_dim=1,
+            min_positive=0.4,
+            max_positive=0.6,
+            min_abs=1.0,
+            max_abs=6.0,
+        )
+        self.out_whiten = Whiten(
+            num_groups=1,
+            whitening_limit=5.0,
+            prob=(0.025, 0.25),
+            grad_scale=0.01,
+        )
+
+    def forward(self, x: Tensor) -> Tensor:
+        if torch.jit.is_scripting() or torch.jit.is_tracing() or not self.training:
+            return self.forward_internal(x)
+        layerdrop_rate = float(self.layerdrop_rate)
+
+        if layerdrop_rate != 0.0:
+            batch_size = x.shape[0]
+            mask = (
+                torch.rand((batch_size, 1, 1, 1), dtype=x.dtype, device=x.device)
+                > layerdrop_rate
+            )
+        else:
+            mask = None
+        # turns out this caching idea does not work with --world-size > 1
+        # return caching_eval(self.forward_internal, x, mask)
+        return self.forward_internal(x, mask)
+
+    def forward_internal(
+        self, x: Tensor, layer_skip_mask: Optional[Tensor] = None
+    ) -> Tensor:
+        """
+        x layout: (N, C, H, W), i.e. (batch_size, num_channels, num_frames, num_freqs)
+
+        The returned value has the same shape as x.
+        """
+        bypass = x
+        x = self.depthwise_conv(x)
+        x = self.pointwise_conv1(x)
+        x = self.hidden_balancer(x)
+        x = self.activation(x)
+        x = self.pointwise_conv2(x)
+
+        if layer_skip_mask is not None:
+            x = x * layer_skip_mask
+
+        x = bypass + x
+        x = self.out_balancer(x)
+
+        if x.requires_grad:
+            x = x.transpose(1, 3)  # (N, W, H, C); need channel dim to be last
+            x = self.out_whiten(x)
+            x = x.transpose(1, 3)  # (N, C, H, W)
+
+        return x
+
+    def streaming_forward(
+        self,
+        x: Tensor,
+        cached_left_pad: Tensor,
+    ) -> Tuple[Tensor, Tensor]:
+        """
+        Args:
+            x layout: (N, C, H, W), i.e. (batch_size, num_channels, num_frames, num_freqs)
+            cached_left_pad: (batch_size, num_channels, left_pad, num_freqs)
+
+        Returns:
+            - The returned value has the same shape as x.
+            - Updated cached_left_pad.
+        """
+        padding = self.padding
+
+        # The length without right padding for depth-wise conv
+        T = x.size(2) - padding[0]
+
+        bypass = x[:, :, :T, :]
+
+        # Pad left side
+        assert cached_left_pad.size(2) == padding[0], (
+            cached_left_pad.size(2),
+            padding[0],
+        )
+        x = torch.cat([cached_left_pad, x], dim=2)
+        # Update cached left padding
+        cached_left_pad = x[:, :, T : padding[0] + T, :]
+
+        # depthwise_conv
+        x = torch.nn.functional.conv2d(
+            x,
+            weight=self.depthwise_conv.weight,
+            bias=self.depthwise_conv.bias,
+            padding=(0, padding[1]),
+            groups=self.depthwise_conv.groups,
+        )
+        x = self.pointwise_conv1(x)
+        x = self.hidden_balancer(x)
+        x = self.activation(x)
+        x = self.pointwise_conv2(x)
+
+        x = bypass + x
+        return x, cached_left_pad
+
+
+class Conv2dSubsampling(nn.Module):
+    """Convolutional 2D subsampling (to 1/2 length).
+
+    Convert an input of shape (N, T, idim) to an output
+    with shape (N, T', odim), where
+    T' = (T-3)//2 - 2 == (T-7)//2
+
+    It is based on
+    https://github.com/espnet/espnet/blob/master/espnet/nets/pytorch_backend/transformer/subsampling.py  # noqa
+    """
+
+    def __init__(
+        self,
+        in_channels: int,
+        out_channels: int,
+        layer1_channels: int = 8,
+        layer2_channels: int = 32,
+        layer3_channels: int = 128,
+        dropout: FloatLike = 0.1,
+    ) -> None:
+        """
+        Args:
+          in_channels:
+            Number of channels in. The input shape is (N, T, in_channels).
+            Caution: It requires: T >=7, in_channels >=7
+          out_channels
+            Output dim. The output shape is (N, (T-3)//2, out_channels)
+          layer1_channels:
+            Number of channels in layer1
+          layer1_channels:
+            Number of channels in layer2
+          bottleneck:
+            bottleneck dimension for 1d squeeze-excite
+        """
+        assert in_channels >= 7
+        super().__init__()
+
+        # The ScaleGrad module is there to prevent the gradients
+        # w.r.t. the weight or bias of the first Conv2d module in self.conv from
+        # exceeding the range of fp16 when using automatic mixed precision (amp)
+        # training.  (The second one is necessary to stop its bias from getting
+        # a too-large gradient).
+
+        self.conv = nn.Sequential(
+            nn.Conv2d(
+                in_channels=1,
+                out_channels=layer1_channels,
+                kernel_size=3,
+                padding=(0, 1),  # (time, freq)
+            ),
+            ScaleGrad(0.2),
+            Balancer(layer1_channels, channel_dim=1, max_abs=1.0),
+            SwooshR(),
+            nn.Conv2d(
+                in_channels=layer1_channels,
+                out_channels=layer2_channels,
+                kernel_size=3,
+                stride=2,
+                padding=0,
+            ),
+            Balancer(layer2_channels, channel_dim=1, max_abs=4.0),
+            SwooshR(),
+            nn.Conv2d(
+                in_channels=layer2_channels,
+                out_channels=layer3_channels,
+                kernel_size=3,
+                stride=(1, 2),  # (time, freq)
+            ),
+            Balancer(layer3_channels, channel_dim=1, max_abs=4.0),
+            SwooshR(),
+        )
+
+        # just one convnext layer
+        self.convnext = ConvNeXt(layer3_channels, kernel_size=(7, 7))
+
+        # (in_channels-3)//4
+        self.out_width = (((in_channels - 1) // 2) - 1) // 2
+        self.layer3_channels = layer3_channels
+
+        self.out = nn.Linear(self.out_width * layer3_channels, out_channels)
+        # use a larger than normal grad_scale on this whitening module; there is
+        # only one such module, so there is not a concern about adding together
+        # many copies of this extra gradient term.
+        self.out_whiten = Whiten(
+            num_groups=1,
+            whitening_limit=ScheduledFloat((0.0, 4.0), (20000.0, 8.0), default=4.0),
+            prob=(0.025, 0.25),
+            grad_scale=0.02,
+        )
+
+        # max_log_eps=0.0 is to prevent both eps and the output of self.out from
+        # getting large, there is an unnecessary degree of freedom.
+        self.out_norm = BiasNorm(out_channels)
+        self.dropout = Dropout3(dropout, shared_dim=1)
+
+    def forward(
+        self, x: torch.Tensor, x_lens: torch.Tensor
+    ) -> Tuple[torch.Tensor, torch.Tensor]:
+        """Subsample x.
+
+        Args:
+          x:
+            Its shape is (N, T, idim).
+          x_lens:
+            A tensor of shape (batch_size,) containing the number of frames in
+
+        Returns:
+          - a tensor of shape (N, (T-7)//2, odim)
+          - output lengths, of shape (batch_size,)
+        """
+        # On entry, x is (N, T, idim)
+        x = x.unsqueeze(1)  # (N, T, idim) -> (N, 1, T, idim) i.e., (N, C, H, W)
+        # scaling x by 0.1 allows us to use a larger grad-scale in fp16 "amp" (automatic mixed precision)
+        # training, since the weights in the first convolution are otherwise the limiting factor for getting infinite
+        # gradients.
+        x = self.conv(x)
+        x = self.convnext(x)
+
+        # Now x is of shape (N, odim, (T-7)//2, (idim-3)//4)
+        b, c, t, f = x.size()
+
+        x = x.transpose(1, 2).reshape(b, t, c * f)
+        # now x: (N, (T-7)//2, out_width * layer3_channels))
+
+        x = self.out(x)
+        # Now x is of shape (N, (T-7)//2, odim)
+        x = self.out_whiten(x)
+        x = self.out_norm(x)
+        x = self.dropout(x)
+
+        if torch.jit.is_scripting() or torch.jit.is_tracing():
+            x_lens = (x_lens - 7) // 2
+        else:
+            with warnings.catch_warnings():
+                warnings.simplefilter("ignore")
+                x_lens = (x_lens - 7) // 2
+        assert x.size(1) == x_lens.max().item(), (x.size(1), x_lens.max())
+
+        return x, x_lens
+
+    def streaming_forward(
+        self,
+        x: torch.Tensor,
+        x_lens: torch.Tensor,
+        cached_left_pad: Tensor,
+    ) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor]:
+        """Subsample x.
+
+        Args:
+          x:
+            Its shape is (N, T, idim).
+          x_lens:
+            A tensor of shape (batch_size,) containing the number of frames in
+
+        Returns:
+          - a tensor of shape (N, (T-7)//2, odim)
+          - output lengths, of shape (batch_size,)
+          - updated cache
+        """
+        # On entry, x is (N, T, idim)
+        x = x.unsqueeze(1)  # (N, T, idim) -> (N, 1, T, idim) i.e., (N, C, H, W)
+
+        # T' = (T-7)//2
+        x = self.conv(x)
+
+        # T' = (T-7)//2-3
+        x, cached_left_pad = self.convnext.streaming_forward(
+            x, cached_left_pad=cached_left_pad
+        )
+
+        # Now x is of shape (N, odim, T', ((idim-1)//2 - 1)//2)
+        b, c, t, f = x.size()
+
+        x = x.transpose(1, 2).reshape(b, t, c * f)
+        # now x: (N, T', out_width * layer3_channels))
+
+        x = self.out(x)
+        # Now x is of shape (N, T', odim)
+        x = self.out_norm(x)
+
+        if torch.jit.is_scripting() or torch.jit.is_tracing():
+            assert self.convnext.padding[0] == 3
+            # The ConvNeXt module needs 3 frames of right padding after subsampling
+            x_lens = (x_lens - 7) // 2 - 3
+        else:
+            with warnings.catch_warnings():
+                warnings.simplefilter("ignore")
+                # The ConvNeXt module needs 3 frames of right padding after subsampling
+                assert self.convnext.padding[0] == 3
+                x_lens = (x_lens - 7) // 2 - 3
+
+        assert x.size(1) == x_lens.max().item(), (x.shape, x_lens.max())
+
+        return x, x_lens, cached_left_pad
+
+    @torch.jit.export
+    def get_init_states(
+        self,
+        batch_size: int = 1,
+        device: torch.device = torch.device("cpu"),
+    ) -> Tensor:
+        """Get initial states for Conv2dSubsampling module.
+        It is the cached left padding for ConvNeXt module,
+        of shape (batch_size, num_channels, left_pad, num_freqs)
+        """
+        left_pad = self.convnext.padding[0]
+        freq = self.out_width
+        channels = self.layer3_channels
+        cached_embed_left_pad = torch.zeros(batch_size, channels, left_pad, freq).to(
+            device
+        )
+
+        return cached_embed_left_pad