Zipformer recipe for Cantonese dataset MDCC (#1537)

* init commit

* Create README.md

* handle code switching cases

* misc. fixes

* added manifest statistics

* init commit for the zipformer recipe

* added scripts for exporting model

* added RESULTS.md

* added scripts for streaming related stuff

* doc str fixed

egs/aishell/ASR/README.md

@@ -19,7 +19,9 @@ The following table lists the differences among them.
 | `transducer_stateless_modified`    | Conformer | Embedding + Conv1d | with modified transducer from `optimized_transducer`                     |
 | `transducer_stateless_modified-2`  | Conformer | Embedding + Conv1d | with modified transducer from `optimized_transducer` + extra data      |
 | `pruned_transducer_stateless3`     | Conformer (reworked) | Embedding + Conv1d | pruned RNN-T + reworked model with random combiner + using aidatatang_20zh as extra data|
-| `pruned_transducer_stateless7`     | Zipformer | Embedding | pruned RNN-T + zipformer encoder + stateless decoder with context-size 1 |
+| `pruned_transducer_stateless7`     | Zipformer | Embedding | pruned RNN-T + zipformer encoder + stateless decoder with context-size set to 1 |
+| `zipformer`                           | Upgraded Zipformer | Embedding + Conv1d | The latest recipe with context-size set to 1 |
+
 
 The decoder in `transducer_stateless` is modified from the paper
 [Rnn-Transducer with Stateless Prediction Network](https://ieeexplore.ieee.org/document/9054419/).

egs/aishell/ASR/prepare.sh

@@ -360,7 +360,7 @@ if [ $stage -le 11 ] && [ $stop_stage -ge 11 ]; then
 fi
 
 if [ $stage -le 12 ] && [ $stop_stage -ge 12 ]; then
-  log "Stage 11: Train RNN LM model"
+  log "Stage 12: Train RNN LM model"
   python ../../../icefall/rnn_lm/train.py \
     --start-epoch 0 \
     --world-size 1 \

egs/mdcc/ASR/README.md

@@ -0,0 +1,19 @@
+# Introduction
+
+Multi-Domain Cantonese Corpus (MDCC), consists of 73.6 hours of clean read speech paired with 
+transcripts, collected from Cantonese audiobooks from Hong Kong. It comprises philosophy, 
+politics, education, culture, lifestyle and family domains, covering a wide range of topics. 
+
+Manuscript can be found at: https://arxiv.org/abs/2201.02419
+
+# Transducers
+
+
+
+|                                       | Encoder             | Decoder            | Comment                     |
+|---------------------------------------|---------------------|--------------------|-----------------------------|
+| `zipformer`                           | Upgraded Zipformer | Embedding + Conv1d | The latest recipe with context-size set to 1 |
+
+The decoder is modified from the paper
+[Rnn-Transducer with Stateless Prediction Network](https://ieeexplore.ieee.org/document/9054419/).
+We place an additional Conv1d layer right after the input embedding layer.

egs/mdcc/ASR/RESULTS.md

@@ -0,0 +1,41 @@
+## Results
+
+#### Zipformer
+
+See <https://github.com/k2-fsa/icefall/pull/1537>
+
+[./zipformer](./zipformer)
+
+##### normal-scaled model, number of model parameters: 74470867, i.e., 74.47 M
+
+|                        | test | valid | comment                                 |
+|------------------------|------|-------|-----------------------------------------|
+| greedy search          | 7.45 | 7.51  | --epoch 45 --avg 35                     |
+| modified beam search   | 6.68 | 6.73  | --epoch 45 --avg 35                     |
+| fast beam search       | 7.22 | 7.28  | --epoch 45 --avg 35                     |
+
+The training command:
+
+```
+export CUDA_VISIBLE_DEVICES="0,1,2,3"
+
+./zipformer/train.py \
+  --world-size 4 \
+  --start-epoch 1 \
+  --num-epochs 50 \
+  --use-fp16 1 \
+  --exp-dir ./zipformer/exp \
+  --max-duration 1000 
+```
+
+The decoding command:
+
+```
+ ./zipformer/decode.py \
+   --epoch 45 \
+   --avg 35 \
+   --exp-dir ./zipformer/exp \
+   --decoding-method greedy_search  # modified_beam_search
+```
+
+The pretrained model is available at:  https://huggingface.co/zrjin/icefall-asr-mdcc-zipformer-2024-03-11/

egs/mdcc/ASR/local/compile_hlg.py

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

egs/mdcc/ASR/local/compile_hlg_using_openfst.py

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

egs/mdcc/ASR/local/compile_lg.py

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

egs/mdcc/ASR/local/compute_fbank_mdcc.py

@@ -0,0 +1,157 @@
+#!/usr/bin/env python3
+# Copyright    2021-2024  Xiaomi Corp.   (authors: Fangjun Kuang,
+#                                                  Zengrui Jin,)
+#
+# 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 of the aishell dataset.
+It looks for manifests in the directory data/manifests.
+
+The generated fbank features are saved in data/fbank.
+"""
+
+import argparse
+import logging
+import os
+from pathlib import Path
+
+import torch
+from lhotse import (
+    CutSet,
+    Fbank,
+    FbankConfig,
+    LilcomChunkyWriter,
+    WhisperFbank,
+    WhisperFbankConfig,
+)
+from lhotse.recipes.utils import read_manifests_if_cached
+
+from icefall.utils import get_executor, str2bool
+
+# 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).
+torch.set_num_threads(1)
+torch.set_num_interop_threads(1)
+
+
+def compute_fbank_mdcc(
+    num_mel_bins: int = 80,
+    perturb_speed: bool = False,
+    whisper_fbank: bool = False,
+    output_dir: str = "data/fbank",
+):
+    src_dir = Path("data/manifests")
+    output_dir = Path(output_dir)
+    num_jobs = min(15, os.cpu_count())
+
+    dataset_parts = (
+        "train",
+        "valid",
+        "test",
+    )
+    prefix = "mdcc"
+    suffix = "jsonl.gz"
+    manifests = read_manifests_if_cached(
+        dataset_parts=dataset_parts,
+        output_dir=src_dir,
+        prefix=prefix,
+        suffix=suffix,
+    )
+    assert manifests is not None
+
+    assert len(manifests) == len(dataset_parts), (
+        len(manifests),
+        len(dataset_parts),
+        list(manifests.keys()),
+        dataset_parts,
+    )
+    if whisper_fbank:
+        extractor = WhisperFbank(
+            WhisperFbankConfig(num_filters=num_mel_bins, device="cuda")
+        )
+    else:
+        extractor = Fbank(FbankConfig(num_mel_bins=num_mel_bins))
+
+    with get_executor() as ex:  # Initialize the executor only once.
+        for partition, m in manifests.items():
+            if (output_dir / f"{prefix}_cuts_{partition}.{suffix}").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"],
+            )
+            if "train" in partition and perturb_speed:
+                logging.info("Doing speed perturb")
+                cut_set = (
+                    cut_set + cut_set.perturb_speed(0.9) + cut_set.perturb_speed(1.1)
+                )
+            cut_set = cut_set.compute_and_store_features(
+                extractor=extractor,
+                storage_path=f"{output_dir}/{prefix}_feats_{partition}",
+                # when an executor is specified, make more partitions
+                num_jobs=num_jobs if ex is None else 80,
+                executor=ex,
+                storage_type=LilcomChunkyWriter,
+            )
+            cut_set.to_file(output_dir / f"{prefix}_cuts_{partition}.{suffix}")
+
+
+def get_args():
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "--num-mel-bins",
+        type=int,
+        default=80,
+        help="""The number of mel bins for Fbank""",
+    )
+    parser.add_argument(
+        "--perturb-speed",
+        type=str2bool,
+        default=False,
+        help="Enable 0.9 and 1.1 speed perturbation for data augmentation. Default: False.",
+    )
+    parser.add_argument(
+        "--whisper-fbank",
+        type=str2bool,
+        default=False,
+        help="Use WhisperFbank instead of Fbank. Default: False.",
+    )
+    parser.add_argument(
+        "--output-dir",
+        type=str,
+        default="data/fbank",
+        help="Output directory. Default: data/fbank.",
+    )
+    return parser.parse_args()
+
+
+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_mdcc(
+        num_mel_bins=args.num_mel_bins,
+        perturb_speed=args.perturb_speed,
+        whisper_fbank=args.whisper_fbank,
+        output_dir=args.output_dir,
+    )

egs/mdcc/ASR/local/display_manifest_statistics.py

@@ -0,0 +1,144 @@
+#!/usr/bin/env python3
+# Copyright    2021-2024  Xiaomi Corp.        (authors: Fangjun Kuang,
+#                                                       Zengrui Jin,)
+#
+# 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 displays duration statistics of utterances in a manifest.
+You can use the displayed value to choose minimum/maximum duration
+to remove short and long utterances during the training.
+
+See the function `remove_short_and_long_utt()` in transducer/train.py
+for usage.
+"""
+
+
+from lhotse import load_manifest_lazy
+
+
+def main():
+    path = "./data/fbank/mdcc_cuts_train.jsonl.gz"
+    path = "./data/fbank/mdcc_cuts_valid.jsonl.gz"
+    path = "./data/fbank/mdcc_cuts_test.jsonl.gz"
+
+    cuts = load_manifest_lazy(path)
+    cuts.describe(full=True)
+
+
+if __name__ == "__main__":
+    main()
+
+"""
+data/fbank/mdcc_cuts_train.jsonl.gz (with speed perturbation)
+_________________________________________ 
+_ Cuts count:               _ 195360
+_________________________________________            
+_ Total duration (hh:mm:ss) _ 173:44:59
+_________________________________________               
+_ mean                      _ 3.2
+_________________________________________
+_ std                       _ 2.1
+_________________________________________               
+_ min                       _ 0.2
+_________________________________________
+_ 25%                       _ 1.8        
+_________________________________________
+_ 50%                       _ 2.7
+_________________________________________
+_ 75%                       _ 4.0
+_________________________________________
+_ 99%                       _ 11.0      _
+_________________________________________
+_ 99.5%                     _ 12.4      _
+_________________________________________
+_ 99.9%                     _ 14.8      _
+_________________________________________
+_ max                       _ 16.7      _
+_________________________________________
+_ Recordings available:     _ 195360    _
+_________________________________________
+_ Features available:       _ 195360    _
+_________________________________________
+_ Supervisions available:   _ 195360    _
+_________________________________________
+
+data/fbank/mdcc_cuts_valid.jsonl.gz 
+________________________________________ 
+_ Cuts count:               _ 5663     _ 
+________________________________________ 
+_ Total duration (hh:mm:ss) _ 05:03:12 _ 
+________________________________________ 
+_ mean                      _ 3.2      _ 
+________________________________________ 
+_ std                       _ 2.0      _ 
+________________________________________ 
+_ min                       _ 0.3      _ 
+________________________________________ 
+_ 25%                       _ 1.8      _ 
+________________________________________ 
+_ 50%                       _ 2.7      _ 
+________________________________________ 
+_ 75%                       _ 4.0      _ 
+________________________________________ 
+_ 99%                       _ 10.9     _ 
+________________________________________
+_ 99.5%                     _ 12.3     _
+________________________________________
+_ 99.9%                     _ 14.4     _
+________________________________________
+_ max                       _ 14.8     _
+________________________________________
+_ Recordings available:     _ 5663     _
+________________________________________
+_ Features available:       _ 5663     _
+________________________________________
+_ Supervisions available:   _ 5663     _
+________________________________________
+
+data/fbank/mdcc_cuts_test.jsonl.gz
+________________________________________ 
+_ Cuts count:               _ 12492    _ 
+________________________________________ 
+_ Total duration (hh:mm:ss) _ 11:00:31 _ 
+________________________________________ 
+_ mean                      _ 3.2      _ 
+________________________________________ 
+_ std                       _ 2.0      _ 
+________________________________________ 
+_ min                       _ 0.2      _ 
+________________________________________ 
+_ 25%                       _ 1.8      _ 
+________________________________________ 
+_ 50%                       _ 2.7      _ 
+________________________________________ 
+_ 75%                       _ 4.0      _ 
+________________________________________ 
+_ 99%                       _ 10.5     _ 
+________________________________________ 
+_ 99.5%                     _ 12.1     _ 
+________________________________________
+_ 99.9%                     _ 14.0     _
+________________________________________
+_ max                       _ 14.8     _
+________________________________________
+_ Recordings available:     _ 12492    _
+________________________________________
+_ Features available:       _ 12492    _
+________________________________________
+_ Supervisions available:   _ 12492    _
+________________________________________
+
+"""

egs/mdcc/ASR/local/prepare_char.py

@@ -0,0 +1 @@
+../../../aishell/ASR/local/prepare_char.py

egs/mdcc/ASR/local/prepare_char_lm_training_data.py

@@ -0,0 +1 @@
+../../../aishell/ASR/local/prepare_char_lm_training_data.py

egs/mdcc/ASR/local/prepare_lang.py

@@ -0,0 +1 @@
+../../../aishell/ASR/local/prepare_lang.py

egs/mdcc/ASR/local/prepare_lang_fst.py

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

egs/mdcc/ASR/local/preprocess_mdcc.py

@@ -0,0 +1,157 @@
+#!/usr/bin/env python3
+# Copyright    2024  Xiaomi Corp.        (authors: Zengrui Jin)
+#
+# 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 takes a text file "data/lang_char/text" as input, the file consist of
+lines each containing a transcript, applies text norm and generates the following
+files in the directory "data/lang_char":
+    - text_norm
+    - words.txt
+    - words_no_ids.txt
+    - text_words_segmentation
+"""
+
+import argparse
+import logging
+from pathlib import Path
+from typing import List
+
+import pycantonese
+from tqdm.auto import tqdm
+
+from icefall.utils import is_cjk
+
+
+def get_parser():
+    parser = argparse.ArgumentParser(
+        description="Prepare char lexicon",
+        formatter_class=argparse.ArgumentDefaultsHelpFormatter,
+    )
+    parser.add_argument(
+        "--input-file",
+        "-i",
+        default="data/lang_char/text",
+        type=str,
+        help="The input text file",
+    )
+    parser.add_argument(
+        "--output-dir",
+        "-o",
+        default="data/lang_char",
+        type=str,
+        help="The output directory",
+    )
+    return parser
+
+
+def get_norm_lines(lines: List[str]) -> List[str]:
+    def _text_norm(text: str) -> str:
+        # to cope with the protocol for transcription:
+        # When taking notes, the annotators adhere to the following guidelines:
+        # 1) If the audio contains pure music, the annotators mark the label
+        # "(music)" in the file name of its transcript. 2) If the utterance
+        # contains one or several sentences with background music or noise, the
+        # annotators mark the label "(music)" before each sentence in the transcript.
+        # 3) The annotators use {} symbols to enclose words they are uncertain
+        # about, for example, {梁佳佳}，我是{}人.
+
+        # here we manually fix some errors in the transcript
+
+        return (
+            text.strip()
+            .replace("(music)", "")
+            .replace("(music", "")
+            .replace("{", "")
+            .replace("}", "")
+            .replace("BB所以就指腹為親喇", "BB 所以就指腹為親喇")
+            .upper()
+        )
+
+    return [_text_norm(line) for line in lines]
+
+
+def get_word_segments(lines: List[str]) -> List[str]:
+    # the current pycantonese segmenter does not handle the case when the input
+    # is code switching, so we need to handle it separately
+
+    new_lines = []
+
+    for line in tqdm(lines, desc="Segmenting lines"):
+        try:
+            # code switching
+            if len(line.strip().split(" ")) > 1:
+                segments = []
+                for segment in line.strip().split(" "):
+                    if segment.strip() == "":
+                        continue
+                    try:
+                        if not is_cjk(segment[0]):  # en segment
+                            segments.append(segment)
+                        else:  # zh segment
+                            segments.extend(pycantonese.segment(segment))
+                    except Exception as e:
+                        logging.error(f"Failed to process segment: {segment}")
+                        raise e
+                new_lines.append(" ".join(segments) + "\n")
+            # not code switching
+            else:
+                new_lines.append(" ".join(pycantonese.segment(line)) + "\n")
+        except Exception as e:
+            logging.error(f"Failed to process line: {line}")
+            raise e
+    return new_lines
+
+
+def get_words(lines: List[str]) -> List[str]:
+    words = set()
+    for line in tqdm(lines, desc="Getting words"):
+        words.update(line.strip().split(" "))
+    return list(words)
+
+
+if __name__ == "__main__":
+    parser = get_parser()
+    args = parser.parse_args()
+
+    input_file = Path(args.input_file)
+    output_dir = Path(args.output_dir)
+
+    assert output_dir.is_dir(), f"{output_dir} does not exist"
+    assert input_file.is_file(), f"{input_file} does not exist"
+
+    lines = input_file.read_text(encoding="utf-8").strip().split("\n")
+
+    norm_lines = get_norm_lines(lines)
+    with open(output_dir / "text_norm", "w+", encoding="utf-8") as f:
+        f.writelines([line + "\n" for line in norm_lines])
+
+    text_words_segments = get_word_segments(norm_lines)
+    with open(output_dir / "text_words_segmentation", "w+", encoding="utf-8") as f:
+        f.writelines(text_words_segments)
+
+    words = get_words(text_words_segments)[1:]  # remove "\n" from words
+    with open(output_dir / "words_no_ids.txt", "w+", encoding="utf-8") as f:
+        f.writelines([word + "\n" for word in sorted(words)])
+
+    words = (
+        ["<eps>", "!SIL", "<SPOKEN_NOISE>", "<UNK>"]
+        + sorted(words)
+        + ["#0", "<s>", "<\s>"]
+    )
+
+    with open(output_dir / "words.txt", "w+", encoding="utf-8") as f:
+        f.writelines([f"{word} {i}\n" for i, word in enumerate(words)])

egs/mdcc/ASR/local/text2segments.py

@@ -0,0 +1,86 @@
+#!/usr/bin/env python
+# -*- coding: utf-8 -*-
+
+# Copyright    2021  Xiaomi Corp.        (authors: Mingshuang Luo)
+#              2022  Xiaomi Corp.        (authors: Weiji Zhuang)
+#              2024  Xiaomi Corp.        (authors: Zengrui Jin)
+#
+# 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 takes as input "text", which refers to the transcript file for
+MDCC:
+    - text
+and generates the output file text_word_segmentation which is implemented
+with word segmenting:
+    - text_words_segmentation
+"""
+
+import argparse
+from typing import List
+
+import pycantonese
+from tqdm.auto import tqdm
+
+
+def get_parser():
+    parser = argparse.ArgumentParser(
+        description="Cantonese Word Segmentation for text",
+        formatter_class=argparse.ArgumentDefaultsHelpFormatter,
+    )
+    parser.add_argument(
+        "--input-file",
+        "-i",
+        default="data/lang_char/text",
+        type=str,
+        help="the input text file for MDCC",
+    )
+    parser.add_argument(
+        "--output-file",
+        "-o",
+        default="data/lang_char/text_words_segmentation",
+        type=str,
+        help="the text implemented with words segmenting for MDCC",
+    )
+
+    return parser
+
+
+def get_word_segments(lines: List[str]) -> List[str]:
+    return [
+        " ".join(pycantonese.segment(line)) + "\n"
+        for line in tqdm(lines, desc="Segmenting lines")
+    ]
+
+
+def main():
+    parser = get_parser()
+    args = parser.parse_args()
+
+    input_file = args.input_file
+    output_file = args.output_file
+
+    with open(input_file, "r", encoding="utf-8") as fr:
+        lines = fr.readlines()
+
+        new_lines = get_word_segments(lines)
+
+    with open(output_file, "w", encoding="utf-8") as fw:
+        fw.writelines(new_lines)
+
+
+if __name__ == "__main__":
+    main()

egs/mdcc/ASR/local/text2token.py

@@ -0,0 +1 @@
+../../../aidatatang_200zh/ASR/local/text2token.py

egs/mdcc/ASR/prepare.sh

@@ -0,0 +1,308 @@
+#!/usr/bin/env bash
+
+# fix segmentation fault reported in https://github.com/k2-fsa/icefall/issues/674
+export PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION=python
+
+set -eou pipefail
+
+stage=-1
+stop_stage=100
+perturb_speed=true
+
+
+# We assume dl_dir (download dir) contains the following
+# directories and files. If not, they will be downloaded
+# by this script automatically.
+#
+#  - $dl_dir/mdcc
+#       |-- README.md
+#       |-- audio/
+#       |-- clip_info_rthk.csv
+#       |-- cnt_asr_metadata_full.csv
+#       |-- cnt_asr_test_metadata.csv
+#       |-- cnt_asr_train_metadata.csv
+#       |-- cnt_asr_valid_metadata.csv
+#       |-- data_statistic.py
+#       |-- length
+#       |-- podcast_447_2021.csv
+#       |-- test.txt
+#       |-- transcription/
+#       `-- words_length
+#      You can download them from:
+#      https://drive.google.com/file/d/1epfYMMhXdBKA6nxPgUugb2Uj4DllSxkn/view?usp=drive_link
+#
+#  - $dl_dir/musan
+#      This directory contains the following directories downloaded from
+#       http://www.openslr.org/17/
+#
+#     - music
+#     - noise
+#     - speech
+
+dl_dir=$PWD/download
+
+. shared/parse_options.sh || exit 1
+
+# 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"
+
+  # If you have pre-downloaded it to /path/to/mdcc,
+  # you can create a symlink
+  #
+  #   ln -sfv /path/to/mdcc $dl_dir/mdcc
+  #
+  # The directory structure is
+  # mdcc/
+  #  |-- README.md
+  #  |-- audio/
+  #  |-- clip_info_rthk.csv
+  #  |-- cnt_asr_metadata_full.csv
+  #  |-- cnt_asr_test_metadata.csv
+  #  |-- cnt_asr_train_metadata.csv
+  #  |-- cnt_asr_valid_metadata.csv
+  #  |-- data_statistic.py
+  #  |-- length
+  #  |-- podcast_447_2021.csv
+  #  |-- test.txt
+  #  |-- transcription/
+  #  `-- words_length
+
+  if [ ! -d $dl_dir/mdcc/audio ]; then
+    lhotse download mdcc $dl_dir
+
+    # this will download and unzip dataset.zip to $dl_dir/
+
+    mv $dl_dir/dataset $dl_dir/mdcc
+  fi
+
+  # If you have pre-downloaded it to /path/to/musan,
+  # you can create a symlink
+  #
+  #   ln -sfv /path/to/musan $dl_dir/musan
+  #
+  if [ ! -d $dl_dir/musan ]; then
+    lhotse download musan $dl_dir
+  fi
+fi
+
+if [ $stage -le 1 ] && [ $stop_stage -ge 1 ]; then
+  log "Stage 1: Prepare MDCC manifest"
+  # We assume that you have downloaded the MDCC corpus
+  # to $dl_dir/mdcc
+  if [ ! -f data/manifests/.mdcc_manifests.done ]; then
+    log "Might take 40 minutes to traverse the directory."
+    mkdir -p data/manifests
+    lhotse prepare mdcc $dl_dir/mdcc data/manifests
+    touch data/manifests/.mdcc_manifests.done
+  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 data/manifests/.musan_manifests.done ]; then
+    log "It may take 6 minutes"
+    mkdir -p data/manifests
+    lhotse prepare musan $dl_dir/musan data/manifests
+    touch data/manifests/.musan_manifests.done
+  fi
+fi
+
+if [ $stage -le 3 ] && [ $stop_stage -ge 3 ]; then
+  log "Stage 3: Compute fbank for MDCC"
+  if [ ! -f data/fbank/.mdcc.done ]; then
+    mkdir -p data/fbank
+    ./local/compute_fbank_mdcc.py --perturb-speed ${perturb_speed}
+    touch data/fbank/.mdcc.done
+  fi
+fi
+
+if [ $stage -le 4 ] && [ $stop_stage -ge 4 ]; then
+  log "Stage 4: Compute fbank for musan"
+  if [ ! -f data/fbank/.msuan.done ]; then
+    mkdir -p data/fbank
+    ./local/compute_fbank_musan.py
+    touch data/fbank/.msuan.done
+  fi
+fi
+
+lang_char_dir=data/lang_char
+if [ $stage -le 5 ] && [ $stop_stage -ge 5 ]; then
+  log "Stage 5: Prepare char based lang"
+  mkdir -p $lang_char_dir
+
+  # Prepare text.
+  # Note: in Linux, you can install jq with the following command:
+  # 1. wget -O jq https://github.com/stedolan/jq/releases/download/jq-1.6/jq-linux64
+  # 2. chmod +x ./jq
+  # 3. cp jq /usr/bin
+  if [ ! -f $lang_char_dir/text ]; then
+    gunzip -c data/manifests/mdcc_supervisions_train.jsonl.gz \
+      |jq '.text' | sed 's/"//g' | ./local/text2token.py -t "char" \
+      > $lang_char_dir/train_text
+
+    cat $lang_char_dir/train_text > $lang_char_dir/text
+
+    gunzip -c data/manifests/mdcc_supervisions_test.jsonl.gz \
+      |jq '.text' | sed 's/"//g' | ./local/text2token.py -t "char" \
+      > $lang_char_dir/valid_text
+    
+    cat $lang_char_dir/valid_text >> $lang_char_dir/text
+  
+    gunzip -c data/manifests/mdcc_supervisions_valid.jsonl.gz \
+      |jq '.text' | sed 's/"//g' | ./local/text2token.py -t "char" \
+      > $lang_char_dir/test_text
+
+    cat $lang_char_dir/test_text >> $lang_char_dir/text
+  fi
+
+  if [ ! -f $lang_char_dir/text_words_segmentation ]; then
+    ./local/preprocess_mdcc.py --input-file $lang_char_dir/text \
+      --output-dir $lang_char_dir
+    
+    mv $lang_char_dir/text $lang_char_dir/_text
+    cp $lang_char_dir/text_words_segmentation $lang_char_dir/text
+  fi
+
+  if [ ! -f $lang_char_dir/tokens.txt ]; then
+    ./local/prepare_char.py --lang-dir $lang_char_dir
+  fi
+fi
+
+if [ $stage -le 6 ] && [ $stop_stage -ge 6 ]; then
+  log "Stage 6: Prepare G"
+
+  mkdir -p data/lm
+
+  # Train LM on transcripts
+  if [ ! -f data/lm/3-gram.unpruned.arpa ]; then
+    python3 ./shared/make_kn_lm.py \
+      -ngram-order 3 \
+      -text $lang_char_dir/text_words_segmentation \
+      -lm data/lm/3-gram.unpruned.arpa
+  fi
+
+  # We assume you have installed kaldilm, if not, please install
+  # it using: pip install kaldilm
+  if [ ! -f data/lm/G_3_gram_char.fst.txt ]; then
+    # It is used in building HLG
+    python3 -m kaldilm \
+      --read-symbol-table="$lang_char_dir/words.txt" \
+      --disambig-symbol='#0' \
+      --max-order=3 \
+      data/lm/3-gram.unpruned.arpa > data/lm/G_3_gram_char.fst.txt
+  fi
+
+  if [ ! -f $lang_char_dir/HLG.fst ]; then
+    ./local/prepare_lang_fst.py  \
+      --lang-dir $lang_char_dir \
+      --ngram-G ./data/lm/G_3_gram_char.fst.txt
+  fi
+fi
+
+if [ $stage -le 7 ] && [ $stop_stage -ge 7 ]; then
+  log "Stage 7: Compile LG & HLG"
+
+  ./local/compile_hlg.py --lang-dir $lang_char_dir --lm G_3_gram_char
+  ./local/compile_lg.py --lang-dir $lang_char_dir --lm G_3_gram_char
+fi
+
+if [ $stage -le 8 ] && [ $stop_stage -ge 8 ]; then
+  log "Stage 8: Generate LM training data"
+
+  log "Processing char based data"
+  out_dir=data/lm_training_char
+  mkdir -p $out_dir $dl_dir/lm
+
+  if [ ! -f $dl_dir/lm/mdcc-train-word.txt ]; then
+    ./local/text2segments.py --input-file $lang_char_dir/train_text \
+        --output-file $dl_dir/lm/mdcc-train-word.txt
+  fi
+
+  # training words
+  ./local/prepare_char_lm_training_data.py \
+    --lang-char data/lang_char \
+    --lm-data $dl_dir/lm/mdcc-train-word.txt \
+    --lm-archive $out_dir/lm_data.pt
+  
+  # valid words
+  if [ ! -f $dl_dir/lm/mdcc-valid-word.txt ]; then
+    ./local/text2segments.py --input-file $lang_char_dir/valid_text \
+        --output-file $dl_dir/lm/mdcc-valid-word.txt
+  fi
+
+  ./local/prepare_char_lm_training_data.py \
+    --lang-char data/lang_char \
+    --lm-data $dl_dir/lm/mdcc-valid-word.txt \
+    --lm-archive $out_dir/lm_data_valid.pt
+
+  # test words
+  if [ ! -f $dl_dir/lm/mdcc-test-word.txt ]; then
+    ./local/text2segments.py --input-file $lang_char_dir/test_text \
+        --output-file $dl_dir/lm/mdcc-test-word.txt
+  fi
+
+  ./local/prepare_char_lm_training_data.py \
+    --lang-char data/lang_char \
+    --lm-data $dl_dir/lm/mdcc-test-word.txt \
+    --lm-archive $out_dir/lm_data_test.pt
+fi
+
+if [ $stage -le 9 ] && [ $stop_stage -ge 9 ]; then
+  log "Stage 9: Sort LM training data"
+  # Sort LM training data by sentence length in descending order
+  # for ease of training.
+  #
+  # Sentence length equals to the number of tokens
+  # in a sentence.
+
+  out_dir=data/lm_training_char
+  mkdir -p $out_dir
+  ln -snf ../../../librispeech/ASR/local/sort_lm_training_data.py local/
+
+  ./local/sort_lm_training_data.py \
+    --in-lm-data $out_dir/lm_data.pt \
+    --out-lm-data $out_dir/sorted_lm_data.pt \
+    --out-statistics $out_dir/statistics.txt
+
+  ./local/sort_lm_training_data.py \
+    --in-lm-data $out_dir/lm_data_valid.pt \
+    --out-lm-data $out_dir/sorted_lm_data-valid.pt \
+    --out-statistics $out_dir/statistics-valid.txt
+
+  ./local/sort_lm_training_data.py \
+    --in-lm-data $out_dir/lm_data_test.pt \
+    --out-lm-data $out_dir/sorted_lm_data-test.pt \
+    --out-statistics $out_dir/statistics-test.txt
+fi
+
+if [ $stage -le 12 ] && [ $stop_stage -ge 12 ]; then
+  log "Stage 12: Train RNN LM model"
+  python ../../../icefall/rnn_lm/train.py \
+    --start-epoch 0 \
+    --world-size 1 \
+    --num-epochs 20 \
+    --use-fp16 0 \
+    --embedding-dim 512 \
+    --hidden-dim 512 \
+    --num-layers 2 \
+    --batch-size 400 \
+    --exp-dir rnnlm_char/exp \
+    --lm-data $out_dir/sorted_lm_data.pt \
+    --lm-data-valid $out_dir/sorted_lm_data-valid.pt \
+    --vocab-size 4336 \
+    --master-port 12345
+fi

egs/mdcc/ASR/shared

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

egs/mdcc/ASR/zipformer/asr_datamodule.py

@@ -0,0 +1,382 @@
+# Copyright      2021  Piotr Żelasko
+# Copyright      2022  Xiaomi Corporation     (Author: Mingshuang Luo)
+# Copyright      2024  Xiaomi Corporation     (Author: Zengrui Jin)
+#
+# 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, List, Optional
+
+from lhotse import CutSet, Fbank, FbankConfig, load_manifest, load_manifest_lazy
+from lhotse.dataset import (
+    CutConcatenate,
+    CutMix,
+    DynamicBucketingSampler,
+    K2SpeechRecognitionDataset,
+    PrecomputedFeatures,
+    SimpleCutSampler,
+    SpecAugment,
+)
+from lhotse.dataset.input_strategies import OnTheFlyFeatures
+from torch.utils.data import DataLoader
+
+from icefall.utils import str2bool
+
+
+class MdccAsrDataModule:
+    """
+    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=2,
+            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. ",
+        )
+
+    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.
+        """
+        logging.info("About to get Musan cuts")
+        cuts_musan = load_manifest(self.args.manifest_dir / "musan_cuts.jsonl.gz")
+
+        transforms = []
+        if self.args.enable_musan:
+            logging.info("Enable MUSAN")
+            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 = K2SpeechRecognitionDataset(
+            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 = K2SpeechRecognitionDataset(
+                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,
+                buffer_size=self.args.num_buckets * 2000,
+                shuffle_buffer_size=self.args.num_buckets * 5000,
+                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)
+
+        train_dl = DataLoader(
+            train,
+            sampler=train_sampler,
+            batch_size=None,
+            num_workers=self.args.num_workers,
+            persistent_workers=False,
+        )
+
+        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 = K2SpeechRecognitionDataset(
+                cut_transforms=transforms,
+                input_strategy=OnTheFlyFeatures(Fbank(FbankConfig(num_mel_bins=80))),
+                return_cuts=self.args.return_cuts,
+            )
+        else:
+            validate = K2SpeechRecognitionDataset(
+                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=2,
+            persistent_workers=False,
+        )
+
+        return valid_dl
+
+    def test_dataloaders(self, cuts: CutSet) -> DataLoader:
+        logging.info("About to create test dataset")
+        test = K2SpeechRecognitionDataset(
+            input_strategy=(
+                OnTheFlyFeatures(Fbank(FbankConfig(num_mel_bins=80)))
+                if self.args.on_the_fly_feats
+                else PrecomputedFeatures()
+            ),
+            return_cuts=self.args.return_cuts,
+        )
+        sampler = DynamicBucketingSampler(
+            cuts,
+            max_duration=self.args.max_duration,
+            shuffle=False,
+        )
+        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("About to get train cuts")
+        cuts_train = load_manifest_lazy(
+            self.args.manifest_dir / "mdcc_cuts_train.jsonl.gz"
+        )
+        return cuts_train
+
+    @lru_cache()
+    def valid_cuts(self) -> CutSet:
+        logging.info("About to get valid cuts")
+        return load_manifest_lazy(self.args.manifest_dir / "mdcc_cuts_valid.jsonl.gz")
+
+    @lru_cache()
+    def test_cuts(self) -> List[CutSet]:
+        logging.info("About to get test cuts")
+        return load_manifest_lazy(self.args.manifest_dir / "mdcc_cuts_test.jsonl.gz")

egs/mdcc/ASR/zipformer/beam_search.py

@@ -0,0 +1 @@
+../../../librispeech/ASR/pruned_transducer_stateless2/beam_search.py

egs/mdcc/ASR/zipformer/decode.py

@@ -0,0 +1,813 @@
+#!/usr/bin/env python3
+#
+# Copyright 2021-2024 Xiaomi Corporation (Author: Fangjun Kuang,
+#                                                 Zengwei Yao,
+#                                                 Mingshuang Luo,
+#                                                 Zengrui Jin,)
+#
+# 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:
+(1) greedy search
+./zipformer/decode.py \
+    --epoch 35 \
+    --avg 15 \
+    --exp-dir ./zipformer/exp \
+    --lang-dir data/lang_char \
+    --max-duration 600 \
+    --decoding-method greedy_search
+
+(2) modified beam search
+./zipformer/decode.py \
+    --epoch 35 \
+    --avg 15 \
+    --exp-dir ./zipformer/exp \
+    --lang-dir data/lang_char \
+    --max-duration 600 \
+    --decoding-method modified_beam_search \
+    --beam-size 4
+
+(3) fast beam search (trivial_graph)
+./zipformer/decode.py \
+    --epoch 35 \
+    --avg 15 \
+    --exp-dir ./zipformer/exp \
+    --lang-dir data/lang_char \
+    --max-duration 600 \
+    --decoding-method fast_beam_search \
+    --beam 20.0 \
+    --max-contexts 8 \
+    --max-states 64
+
+(4) fast beam search (LG)
+./zipformer/decode.py \
+    --epoch 30 \
+    --avg 15 \
+    --exp-dir ./zipformer/exp \
+    --lang-dir data/lang_char \
+    --max-duration 600 \
+    --decoding-method fast_beam_search_LG \
+    --beam 20.0 \
+    --max-contexts 8 \
+    --max-states 64
+
+(5) fast beam search (nbest oracle WER)
+./zipformer/decode.py \
+    --epoch 35 \
+    --avg 15 \
+    --exp-dir ./zipformer/exp \
+    --lang-dir data/lang_char \
+    --max-duration 600 \
+    --decoding-method fast_beam_search_nbest_oracle \
+    --beam 20.0 \
+    --max-contexts 8 \
+    --max-states 64 \
+    --num-paths 200 \
+    --nbest-scale 0.5
+"""
+
+
+import argparse
+import logging
+import math
+from collections import defaultdict
+from pathlib import Path
+from typing import Dict, List, Optional, Tuple
+
+import k2
+import torch
+import torch.nn as nn
+from asr_datamodule import MdccAsrDataModule
+from beam_search import (
+    beam_search,
+    fast_beam_search_nbest_oracle,
+    fast_beam_search_one_best,
+    greedy_search,
+    greedy_search_batch,
+    modified_beam_search,
+)
+from lhotse.cut import Cut
+from train import add_model_arguments, get_model, get_params
+
+from icefall.char_graph_compiler import CharCtcTrainingGraphCompiler
+from icefall.checkpoint import (
+    average_checkpoints,
+    average_checkpoints_with_averaged_model,
+    find_checkpoints,
+    load_checkpoint,
+)
+from icefall.lexicon import Lexicon
+from icefall.utils import (
+    AttributeDict,
+    make_pad_mask,
+    setup_logger,
+    store_transcripts,
+    str2bool,
+    write_error_stats,
+)
+
+LOG_EPS = math.log(1e-10)
+
+
+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=15,
+        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="The experiment dir",
+    )
+
+    parser.add_argument(
+        "--lang-dir",
+        type=Path,
+        default="data/lang_char",
+        help="The lang dir containing word table and LG graph",
+    )
+
+    parser.add_argument(
+        "--decoding-method",
+        type=str,
+        default="greedy_search",
+        help="""Possible values are:
+          - greedy_search
+          - modified_beam_search
+          - fast_beam_search
+          - fast_beam_search_LG
+          - fast_beam_search_nbest_oracle
+        If you use fast_beam_search_LG, you have to specify
+        `--lang-dir`, which should contain `LG.pt`.
+        """,
+    )
+
+    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 --decoding-method is beam_search or
+        modified_beam_search.""",
+    )
+
+    parser.add_argument(
+        "--beam",
+        type=float,
+        default=20.0,
+        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 --decoding-method is fast_beam_search,
+        fast_beam_search, fast_beam_search_LG,
+        and fast_beam_search_nbest_oracle
+        """,
+    )
+
+    parser.add_argument(
+        "--ngram-lm-scale",
+        type=float,
+        default=0.01,
+        help="""
+        Used only when --decoding_method is fast_beam_search_LG.
+        It specifies the scale for n-gram LM scores.
+        """,
+    )
+
+    parser.add_argument(
+        "--ilme-scale",
+        type=float,
+        default=0.2,
+        help="""
+        Used only when --decoding_method is fast_beam_search_LG.
+        It specifies the scale for the internal language model estimation.
+        """,
+    )
+
+    parser.add_argument(
+        "--max-contexts",
+        type=int,
+        default=8,
+        help="""Used only when --decoding-method is
+        fast_beam_search, fast_beam_search, fast_beam_search_LG,
+        and fast_beam_search_nbest_oracle""",
+    )
+
+    parser.add_argument(
+        "--max-states",
+        type=int,
+        default=64,
+        help="""Used only when --decoding-method is
+        fast_beam_search, fast_beam_search, fast_beam_search_LG,
+        and fast_beam_search_nbest_oracle""",
+    )
+
+    parser.add_argument(
+        "--context-size",
+        type=int,
+        default=1,
+        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 --decoding_method is greedy_search""",
+    )
+
+    parser.add_argument(
+        "--num-paths",
+        type=int,
+        default=200,
+        help="""Number of paths for nbest decoding.
+        Used only when the decoding method is fast_beam_search_nbest_oracle""",
+    )
+
+    parser.add_argument(
+        "--nbest-scale",
+        type=float,
+        default=0.5,
+        help="""Scale applied to lattice scores when computing nbest paths.
+        Used only when the decoding method is and fast_beam_search_nbest_oracle""",
+    )
+
+    parser.add_argument(
+        "--blank-penalty",
+        type=float,
+        default=0.0,
+        help="""
+        The penalty applied on blank symbol during decoding.
+        Note: It is a positive value that would be applied to logits like
+        this `logits[:, 0] -= blank_penalty` (suppose logits.shape is
+        [batch_size, vocab] and blank id is 0).
+        """,
+    )
+
+    add_model_arguments(parser)
+
+    return parser
+
+
+def decode_one_batch(
+    params: AttributeDict,
+    model: nn.Module,
+    lexicon: Lexicon,
+    graph_compiler: CharCtcTrainingGraphCompiler,
+    batch: dict,
+    decoding_graph: Optional[k2.Fsa] = None,
+) -> Dict[str, List[List[str]]]:
+    """Decode one batch and return the result in a dict. The dict has the
+    following format:
+
+        - key: It indicates the setting used for decoding. For example,
+               if greedy_search is used, it would be "greedy_search"
+               If beam search with a beam size of 7 is used, it would be
+               "beam_7"
+        - value: It contains the decoding result. `len(value)` equals to
+                 batch size. `value[i]` is the decoding result for the i-th
+                 utterance in the given batch.
+    Args:
+      params:
+        It's the return value of :func:`get_params`.
+      model:
+        The neural model.
+      batch:
+        It is the return value from iterating
+        `lhotse.dataset.K2SpeechRecognitionDataset`. See its documentation
+        for the format of the `batch`.
+      decoding_graph:
+        The decoding graph. Can be either a `k2.trivial_graph` or LG, Used
+        only when --decoding_method is fast_beam_search, fast_beam_search_nbest,
+        fast_beam_search_nbest_oracle, and fast_beam_search_nbest_LG.
+    Returns:
+      Return the decoding result. See above description for the format of
+      the returned dict.
+    """
+    device = next(model.parameters()).device
+    feature = batch["inputs"]
+    assert feature.ndim == 3
+
+    feature = feature.to(device)
+    # at entry, feature is (N, T, C)
+
+    supervisions = batch["supervisions"]
+    feature_lens = supervisions["num_frames"].to(device)
+
+    if params.causal:
+        # this seems to cause insertions at the end of the utterance if used with zipformer.
+        pad_len = 30
+        feature_lens += pad_len
+        feature = torch.nn.functional.pad(
+            feature,
+            pad=(0, 0, 0, pad_len),
+            value=LOG_EPS,
+        )
+
+    x, x_lens = model.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 = model.encoder(x, x_lens, src_key_padding_mask)
+    encoder_out = encoder_out.permute(1, 0, 2)  # (T, N, C) ->(N, T, C)
+
+    hyps = []
+
+    if params.decoding_method == "fast_beam_search":
+        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,
+            blank_penalty=params.blank_penalty,
+        )
+        for i in range(encoder_out.size(0)):
+            hyps.append([lexicon.token_table[idx] for idx in hyp_tokens[i]])
+    elif params.decoding_method == "fast_beam_search_LG":
+        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,
+            blank_penalty=params.blank_penalty,
+            ilme_scale=params.ilme_scale,
+        )
+        for hyp in hyp_tokens:
+            sentence = "".join([lexicon.word_table[i] for i in hyp])
+            hyps.append(list(sentence))
+    elif params.decoding_method == "fast_beam_search_nbest_oracle":
+        hyp_tokens = fast_beam_search_nbest_oracle(
+            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,
+            num_paths=params.num_paths,
+            ref_texts=graph_compiler.texts_to_ids(supervisions["text"]),
+            nbest_scale=params.nbest_scale,
+            blank_penalty=params.blank_penalty,
+        )
+        for i in range(encoder_out.size(0)):
+            hyps.append([lexicon.token_table[idx] for idx in hyp_tokens[i]])
+    elif params.decoding_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,
+            blank_penalty=params.blank_penalty,
+        )
+        for i in range(encoder_out.size(0)):
+            hyps.append([lexicon.token_table[idx] for idx in hyp_tokens[i]])
+    elif params.decoding_method == "modified_beam_search":
+        hyp_tokens = modified_beam_search(
+            model=model,
+            encoder_out=encoder_out,
+            encoder_out_lens=encoder_out_lens,
+            blank_penalty=params.blank_penalty,
+            beam=params.beam_size,
+        )
+        for i in range(encoder_out.size(0)):
+            hyps.append([lexicon.token_table[idx] for idx in hyp_tokens[i]])
+    else:
+        batch_size = encoder_out.size(0)
+
+        for i in range(batch_size):
+            # fmt: off
+            encoder_out_i = encoder_out[i:i + 1, :encoder_out_lens[i]]
+            # fmt: on
+            if params.decoding_method == "greedy_search":
+                hyp = greedy_search(
+                    model=model,
+                    encoder_out=encoder_out_i,
+                    max_sym_per_frame=params.max_sym_per_frame,
+                    blank_penalty=params.blank_penalty,
+                )
+            elif params.decoding_method == "beam_search":
+                hyp = beam_search(
+                    model=model,
+                    encoder_out=encoder_out_i,
+                    beam=params.beam_size,
+                    blank_penalty=params.blank_penalty,
+                )
+            else:
+                raise ValueError(
+                    f"Unsupported decoding method: {params.decoding_method}"
+                )
+            hyps.append([lexicon.token_table[idx] for idx in hyp])
+
+    key = f"blank_penalty_{params.blank_penalty}"
+    if params.decoding_method == "greedy_search":
+        return {"greedy_search_" + key: hyps}
+    elif "fast_beam_search" in params.decoding_method:
+        key += f"_beam_{params.beam}_"
+        key += f"max_contexts_{params.max_contexts}_"
+        key += f"max_states_{params.max_states}"
+        if "nbest" in params.decoding_method:
+            key += f"_num_paths_{params.num_paths}_"
+            key += f"nbest_scale_{params.nbest_scale}"
+        if "LG" in params.decoding_method:
+            key += f"_ilme_scale_{params.ilme_scale}"
+            key += f"_ngram_lm_scale_{params.ngram_lm_scale}"
+
+        return {key: hyps}
+    else:
+        return {f"beam_size_{params.beam_size}_" + key: hyps}
+
+
+def decode_dataset(
+    dl: torch.utils.data.DataLoader,
+    params: AttributeDict,
+    model: nn.Module,
+    lexicon: Lexicon,
+    graph_compiler: CharCtcTrainingGraphCompiler,
+    decoding_graph: Optional[k2.Fsa] = None,
+) -> Dict[str, List[Tuple[List[str], List[str]]]]:
+    """Decode dataset.
+
+    Args:
+      dl:
+        PyTorch's dataloader containing the dataset to decode.
+      params:
+        It is returned by :func:`get_params`.
+      model:
+        The neural model.
+      decoding_graph:
+        The decoding graph. Can be either a `k2.trivial_graph` or LG, Used
+        only when --decoding_method is fast_beam_search, fast_beam_search_nbest,
+        fast_beam_search_nbest_oracle, and fast_beam_search_nbest_LG.
+    Returns:
+      Return a dict, whose key may be "greedy_search" if greedy search
+      is used, or it may be "beam_7" if beam size of 7 is used.
+      Its value is a list of tuples. Each tuple contains two elements:
+      The first is the reference transcript, and the second is the
+      predicted result.
+    """
+    num_cuts = 0
+
+    try:
+        num_batches = len(dl)
+    except TypeError:
+        num_batches = "?"
+
+    if params.decoding_method == "greedy_search":
+        log_interval = 50
+    else:
+        log_interval = 20
+
+    results = defaultdict(list)
+    for batch_idx, batch in enumerate(dl):
+        texts = batch["supervisions"]["text"]
+        texts = [list("".join(text.split())) for text in texts]
+        cut_ids = [cut.id for cut in batch["supervisions"]["cut"]]
+
+        hyps_dict = decode_one_batch(
+            params=params,
+            model=model,
+            lexicon=lexicon,
+            graph_compiler=graph_compiler,
+            decoding_graph=decoding_graph,
+            batch=batch,
+        )
+
+        for name, hyps in hyps_dict.items():
+            this_batch = []
+            assert len(hyps) == len(texts)
+            for cut_id, hyp_words, ref_text in zip(cut_ids, hyps, texts):
+                this_batch.append((cut_id, ref_text, hyp_words))
+
+            results[name].extend(this_batch)
+
+        num_cuts += len(texts)
+
+        if batch_idx % log_interval == 0:
+            batch_str = f"{batch_idx}/{num_batches}"
+
+            logging.info(f"batch {batch_str}, cuts processed until now is {num_cuts}")
+    return results
+
+
+def save_results(
+    params: AttributeDict,
+    test_set_name: str,
+    results_dict: Dict[str, List[Tuple[List[int], List[int]]]],
+):
+    test_set_wers = dict()
+    for key, results in results_dict.items():
+        recog_path = (
+            params.res_dir / f"recogs-{test_set_name}-{key}-{params.suffix}.txt"
+        )
+        results = sorted(results)
+        store_transcripts(filename=recog_path, texts=results)
+        logging.info(f"The transcripts are stored in {recog_path}")
+
+        # The following prints out WERs, per-word error statistics and aligned
+        # ref/hyp pairs.
+        errs_filename = (
+            params.res_dir / f"errs-{test_set_name}-{key}-{params.suffix}.txt"
+        )
+        with open(errs_filename, "w") as f:
+            wer = write_error_stats(
+                f, f"{test_set_name}-{key}", results, enable_log=True
+            )
+            test_set_wers[key] = wer
+
+        logging.info("Wrote detailed error stats to {}".format(errs_filename))
+
+    test_set_wers = sorted(test_set_wers.items(), key=lambda x: x[1])
+    errs_info = (
+        params.res_dir / f"wer-summary-{test_set_name}-{key}-{params.suffix}.txt"
+    )
+    with open(errs_info, "w") as f:
+        print("settings\tWER", file=f)
+        for key, val in test_set_wers:
+            print("{}\t{}".format(key, val), file=f)
+
+    s = "\nFor {}, WER of different settings are:\n".format(test_set_name)
+    note = "\tbest for {}".format(test_set_name)
+    for key, val in test_set_wers:
+        s += "{}\t{}{}\n".format(key, val, note)
+        note = ""
+    logging.info(s)
+
+
+@torch.no_grad()
+def main():
+    parser = get_parser()
+    MdccAsrDataModule.add_arguments(parser)
+    args = parser.parse_args()
+    args.exp_dir = Path(args.exp_dir)
+
+    params = get_params()
+    params.update(vars(args))
+
+    assert params.decoding_method in (
+        "greedy_search",
+        "beam_search",
+        "modified_beam_search",
+        "fast_beam_search",
+        "fast_beam_search_LG",
+        "fast_beam_search_nbest_oracle",
+    )
+    params.res_dir = params.exp_dir / params.decoding_method
+
+    if params.iter > 0:
+        params.suffix = f"iter-{params.iter}-avg-{params.avg}"
+    else:
+        params.suffix = f"epoch-{params.epoch}-avg-{params.avg}"
+
+    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."
+        params.suffix += f"-chunk-{params.chunk_size}"
+        params.suffix += f"-left-context-{params.left_context_frames}"
+
+    if "fast_beam_search" in params.decoding_method:
+        params.suffix += f"-beam-{params.beam}"
+        params.suffix += f"-max-contexts-{params.max_contexts}"
+        params.suffix += f"-max-states-{params.max_states}"
+        if "nbest" in params.decoding_method:
+            params.suffix += f"-nbest-scale-{params.nbest_scale}"
+            params.suffix += f"-num-paths-{params.num_paths}"
+        if "LG" in params.decoding_method:
+            params.suffix += f"_ilme_scale_{params.ilme_scale}"
+            params.suffix += f"-ngram-lm-scale-{params.ngram_lm_scale}"
+    elif "beam_search" in params.decoding_method:
+        params.suffix += f"-{params.decoding_method}-beam-size-{params.beam_size}"
+    else:
+        params.suffix += f"-context-{params.context_size}"
+        params.suffix += f"-max-sym-per-frame-{params.max_sym_per_frame}"
+    params.suffix += f"-blank-penalty-{params.blank_penalty}"
+
+    if params.use_averaged_model:
+        params.suffix += "-use-averaged-model"
+
+    setup_logger(f"{params.res_dir}/log-decode-{params.suffix}")
+    logging.info("Decoding started")
+
+    device = torch.device("cpu")
+    if torch.cuda.is_available():
+        device = torch.device("cuda", 0)
+
+    logging.info(f"Device: {device}")
+
+    lexicon = Lexicon(params.lang_dir)
+    params.blank_id = lexicon.token_table["<blk>"]
+    params.vocab_size = max(lexicon.tokens) + 1
+
+    graph_compiler = CharCtcTrainingGraphCompiler(
+        lexicon=lexicon,
+        device=device,
+    )
+
+    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.to(device)
+            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.to(device)
+            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.to(device)
+            model.load_state_dict(
+                average_checkpoints_with_averaged_model(
+                    filename_start=filename_start,
+                    filename_end=filename_end,
+                    device=device,
+                )
+            )
+        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.to(device)
+            model.load_state_dict(
+                average_checkpoints_with_averaged_model(
+                    filename_start=filename_start,
+                    filename_end=filename_end,
+                    device=device,
+                )
+            )
+
+    model.to(device)
+    model.eval()
+
+    if "fast_beam_search" in params.decoding_method:
+        if "LG" in params.decoding_method:
+            lexicon = Lexicon(params.lang_dir)
+            lg_filename = params.lang_dir / "LG.pt"
+            logging.info(f"Loading {lg_filename}")
+            decoding_graph = k2.Fsa.from_dict(
+                torch.load(lg_filename, map_location=device)
+            )
+            decoding_graph.scores *= params.ngram_lm_scale
+        else:
+            decoding_graph = k2.trivial_graph(params.vocab_size - 1, device=device)
+    else:
+        decoding_graph = None
+
+    num_param = sum([p.numel() for p in model.parameters()])
+    logging.info(f"Number of model parameters: {num_param}")
+
+    # we need cut ids to display recognition results.
+    args.return_cuts = True
+    mdcc = MdccAsrDataModule(args)
+
+    def remove_short_utt(c: Cut):
+        T = ((c.num_frames - 7) // 2 + 1) // 2
+        if T <= 0:
+            logging.warning(
+                f"Exclude cut with ID {c.id} from decoding, num_frames : {c.num_frames}."
+            )
+        return T > 0
+
+    valid_cuts = mdcc.valid_cuts()
+    valid_cuts = valid_cuts.filter(remove_short_utt)
+    valid_dl = mdcc.valid_dataloaders(valid_cuts)
+
+    test_cuts = mdcc.test_cuts()
+    test_cuts = test_cuts.filter(remove_short_utt)
+    test_dl = mdcc.test_dataloaders(test_cuts)
+
+    test_sets = ["valid", "test"]
+    test_dls = [valid_dl, test_dl]
+
+    for test_set, test_dl in zip(test_sets, test_dls):
+        results_dict = decode_dataset(
+            dl=test_dl,
+            params=params,
+            model=model,
+            lexicon=lexicon,
+            graph_compiler=graph_compiler,
+            decoding_graph=decoding_graph,
+        )
+
+        save_results(
+            params=params,
+            test_set_name=test_set,
+            results_dict=results_dict,
+        )
+
+    logging.info("Done!")
+
+
+if __name__ == "__main__":
+    main()

egs/mdcc/ASR/zipformer/decode_stream.py

@@ -0,0 +1 @@
+../../../librispeech/ASR/zipformer/decode_stream.py

egs/mdcc/ASR/zipformer/decoder.py

@@ -0,0 +1 @@
+../../../librispeech/ASR/zipformer/decoder.py

egs/mdcc/ASR/zipformer/encoder_interface.py

@@ -0,0 +1 @@
+../../../librispeech/ASR/zipformer/encoder_interface.py

egs/mdcc/ASR/zipformer/export-onnx-ctc.py

@@ -0,0 +1 @@
+../../../librispeech/ASR/zipformer/export-onnx-ctc.py

egs/mdcc/ASR/zipformer/export-onnx-streaming-ctc.py

@@ -0,0 +1 @@
+../../../librispeech/ASR/zipformer/export-onnx-streaming-ctc.py

egs/mdcc/ASR/zipformer/export-onnx-streaming.py

@@ -0,0 +1 @@
+../../../librispeech/ASR/zipformer/export-onnx-streaming.py

egs/mdcc/ASR/zipformer/export-onnx.py

@@ -0,0 +1 @@
+../../../librispeech/ASR/zipformer/export-onnx.py

egs/mdcc/ASR/zipformer/export.py

@@ -0,0 +1 @@
+../../../librispeech/ASR/zipformer/export.py

egs/mdcc/ASR/zipformer/joiner.py

@@ -0,0 +1 @@
+../../../librispeech/ASR/zipformer/joiner.py

egs/mdcc/ASR/zipformer/model.py

@@ -0,0 +1 @@
+../../../librispeech/ASR/zipformer/model.py

egs/mdcc/ASR/zipformer/onnx_check.py

@@ -0,0 +1 @@
+../../../librispeech/ASR/zipformer/onnx_check.py

egs/mdcc/ASR/zipformer/onnx_decode.py

@@ -0,0 +1,286 @@
+#!/usr/bin/env python3
+#
+# Copyright 2021-2023 Xiaomi Corporation (Author: Fangjun Kuang,
+#                                                 Zengwei Yao,
+#                                                 Xiaoyu Yang,
+#                                                 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 loads ONNX exported models and uses them to decode the test sets.
+"""
+
+import argparse
+import logging
+import time
+from pathlib import Path
+from typing import List, Tuple
+
+import k2
+import torch
+import torch.nn as nn
+from asr_datamodule import MdccAsrDataModule
+from lhotse.cut import Cut
+from onnx_pretrained import OnnxModel, greedy_search
+
+from icefall.utils import setup_logger, store_transcripts, write_error_stats
+
+
+def get_parser():
+    parser = argparse.ArgumentParser(
+        formatter_class=argparse.ArgumentDefaultsHelpFormatter
+    )
+
+    parser.add_argument(
+        "--encoder-model-filename",
+        type=str,
+        required=True,
+        help="Path to the encoder onnx model. ",
+    )
+
+    parser.add_argument(
+        "--decoder-model-filename",
+        type=str,
+        required=True,
+        help="Path to the decoder onnx model. ",
+    )
+
+    parser.add_argument(
+        "--joiner-model-filename",
+        type=str,
+        required=True,
+        help="Path to the joiner onnx model. ",
+    )
+
+    parser.add_argument(
+        "--exp-dir",
+        type=str,
+        default="pruned_transducer_stateless7/exp",
+        help="The experiment dir",
+    )
+
+    parser.add_argument(
+        "--tokens",
+        type=str,
+        default="data/lang_char/tokens.txt",
+        help="Path to the tokens.txt",
+    )
+
+    parser.add_argument(
+        "--decoding-method",
+        type=str,
+        default="greedy_search",
+        help="Valid values are greedy_search and modified_beam_search",
+    )
+
+    return parser
+
+
+def decode_one_batch(
+    model: OnnxModel, token_table: k2.SymbolTable, batch: dict
+) -> List[List[str]]:
+    """Decode one batch and return the result.
+    Currently it only greedy_search is supported.
+
+    Args:
+      model:
+        The neural model.
+      token_table:
+        Mapping ids to tokens.
+      batch:
+        It is the return value from iterating
+        `lhotse.dataset.K2SpeechRecognitionDataset`. See its documentation
+        for the format of the `batch`.
+
+    Returns:
+      Return the decoded results for each utterance.
+    """
+    feature = batch["inputs"]
+    assert feature.ndim == 3
+    # at entry, feature is (N, T, C)
+
+    supervisions = batch["supervisions"]
+    feature_lens = supervisions["num_frames"].to(dtype=torch.int64)
+
+    encoder_out, encoder_out_lens = model.run_encoder(x=feature, x_lens=feature_lens)
+
+    hyps = greedy_search(
+        model=model, encoder_out=encoder_out, encoder_out_lens=encoder_out_lens
+    )
+
+    hyps = [[token_table[h] for h in hyp] for hyp in hyps]
+    return hyps
+
+
+def decode_dataset(
+    dl: torch.utils.data.DataLoader,
+    model: nn.Module,
+    token_table: k2.SymbolTable,
+) -> Tuple[List[Tuple[str, List[str], List[str]]], float]:
+    """Decode dataset.
+
+    Args:
+      dl:
+        PyTorch's dataloader containing the dataset to decode.
+      model:
+        The neural model.
+      token_table:
+        Mapping ids to tokens.
+
+    Returns:
+      - A list of tuples. Each tuple contains three elements:
+         - cut_id,
+         - reference transcript,
+         - predicted result.
+      - The total duration (in seconds) of the dataset.
+    """
+    num_cuts = 0
+
+    try:
+        num_batches = len(dl)
+    except TypeError:
+        num_batches = "?"
+
+    log_interval = 10
+    total_duration = 0
+
+    results = []
+    for batch_idx, batch in enumerate(dl):
+        texts = batch["supervisions"]["text"]
+        cut_ids = [cut.id for cut in batch["supervisions"]["cut"]]
+        total_duration += sum([cut.duration for cut in batch["supervisions"]["cut"]])
+
+        hyps = decode_one_batch(model=model, token_table=token_table, batch=batch)
+
+        this_batch = []
+        assert len(hyps) == len(texts)
+        for cut_id, hyp_words, ref_text in zip(cut_ids, hyps, texts):
+            ref_words = list(ref_text)
+            this_batch.append((cut_id, ref_words, hyp_words))
+
+        results.extend(this_batch)
+
+        num_cuts += len(texts)
+
+        if batch_idx % log_interval == 0:
+            batch_str = f"{batch_idx}/{num_batches}"
+
+            logging.info(f"batch {batch_str}, cuts processed until now is {num_cuts}")
+
+    return results, total_duration
+
+
+def save_results(
+    res_dir: Path,
+    test_set_name: str,
+    results: List[Tuple[str, List[str], List[str]]],
+):
+    recog_path = res_dir / f"recogs-{test_set_name}.txt"
+    results = sorted(results)
+    store_transcripts(filename=recog_path, texts=results)
+    logging.info(f"The transcripts are stored in {recog_path}")
+
+    # The following prints out WERs, per-word error statistics and aligned
+    # ref/hyp pairs.
+    errs_filename = res_dir / f"errs-{test_set_name}.txt"
+    with open(errs_filename, "w") as f:
+        wer = write_error_stats(f, f"{test_set_name}", results, enable_log=True)
+
+    logging.info("Wrote detailed error stats to {}".format(errs_filename))
+
+    errs_info = res_dir / f"wer-summary-{test_set_name}.txt"
+    with open(errs_info, "w") as f:
+        print("WER", file=f)
+        print(wer, file=f)
+
+    s = "\nFor {}, WER is {}:\n".format(test_set_name, wer)
+    logging.info(s)
+
+
+@torch.no_grad()
+def main():
+    parser = get_parser()
+    MdccAsrDataModule.add_arguments(parser)
+    args = parser.parse_args()
+
+    assert (
+        args.decoding_method == "greedy_search"
+    ), "Only supports greedy_search currently."
+    res_dir = Path(args.exp_dir) / f"onnx-{args.decoding_method}"
+
+    setup_logger(f"{res_dir}/log-decode")
+    logging.info("Decoding started")
+
+    device = torch.device("cpu")
+    logging.info(f"Device: {device}")
+
+    token_table = k2.SymbolTable.from_file(args.tokens)
+    assert token_table[0] == "<blk>"
+
+    logging.info(vars(args))
+
+    logging.info("About to create model")
+    model = OnnxModel(
+        encoder_model_filename=args.encoder_model_filename,
+        decoder_model_filename=args.decoder_model_filename,
+        joiner_model_filename=args.joiner_model_filename,
+    )
+
+    # we need cut ids to display recognition results.
+    args.return_cuts = True
+
+    mdcc = MdccAsrDataModule(args)
+
+    def remove_short_utt(c: Cut):
+        T = ((c.num_frames - 7) // 2 + 1) // 2
+        if T <= 0:
+            logging.warning(
+                f"Exclude cut with ID {c.id} from decoding, num_frames : {c.num_frames}."
+            )
+        return T > 0
+
+    valid_cuts = mdcc.valid_cuts()
+    valid_cuts = valid_cuts.filter(remove_short_utt)
+    valid_dl = mdcc.valid_dataloaders(valid_cuts)
+
+    test_cuts = mdcc.test_net_cuts()
+    test_cuts = test_cuts.filter(remove_short_utt)
+    test_dl = mdcc.test_dataloaders(test_cuts)
+
+    test_sets = ["valid", "test"]
+    test_dl = [valid_dl, test_dl]
+
+    for test_set, test_dl in zip(test_sets, test_dl):
+        start_time = time.time()
+        results, total_duration = decode_dataset(
+            dl=test_dl, model=model, token_table=token_table
+        )
+        end_time = time.time()
+        elapsed_seconds = end_time - start_time
+        rtf = elapsed_seconds / total_duration
+
+        logging.info(f"Elapsed time: {elapsed_seconds:.3f} s")
+        logging.info(f"Wave duration: {total_duration:.3f} s")
+        logging.info(
+            f"Real time factor (RTF): {elapsed_seconds:.3f}/{total_duration:.3f} = {rtf:.3f}"
+        )
+
+        save_results(res_dir=res_dir, test_set_name=test_set, results=results)
+
+    logging.info("Done!")
+
+
+if __name__ == "__main__":
+    main()

egs/mdcc/ASR/zipformer/optim.py

@@ -0,0 +1 @@
+../../../librispeech/ASR/zipformer/optim.py

egs/mdcc/ASR/zipformer/scaling.py

@@ -0,0 +1 @@
+../../../librispeech/ASR/zipformer/scaling.py

egs/mdcc/ASR/zipformer/scaling_converter.py

@@ -0,0 +1 @@
+../../../librispeech/ASR/zipformer/scaling_converter.py

egs/mdcc/ASR/zipformer/streaming_beam_search.py

@@ -0,0 +1 @@
+../../../librispeech/ASR/zipformer/streaming_beam_search.py

egs/mdcc/ASR/zipformer/streaming_decode.py

@@ -0,0 +1,881 @@
+#!/usr/bin/env python3
+# Copyright 2022-2023 Xiaomi Corporation (Authors: Wei Kang,
+#                                                  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.
+
+"""
+Usage:
+./zipformer/streaming_decode.py \
+  --epoch 28 \
+  --avg 15 \
+  --causal 1 \
+  --chunk-size 16 \
+  --left-context-frames 256 \
+  --exp-dir ./zipformer/exp \
+  --decoding-method greedy_search \
+  --num-decode-streams 2000
+"""
+
+import argparse
+import logging
+import math
+from pathlib import Path
+from typing import Dict, List, Optional, Tuple
+
+import k2
+import numpy as np
+import torch
+from asr_datamodule import MdccAsrDataModule
+from decode_stream import DecodeStream
+from kaldifeat import Fbank, FbankOptions
+from lhotse import CutSet
+from streaming_beam_search import (
+    fast_beam_search_one_best,
+    greedy_search,
+    modified_beam_search,
+)
+from torch import Tensor, nn
+from torch.nn.utils.rnn import pad_sequence
+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.lexicon import Lexicon
+from icefall.utils import (
+    AttributeDict,
+    make_pad_mask,
+    setup_logger,
+    store_transcripts,
+    str2bool,
+    write_error_stats,
+)
+
+LOG_EPS = math.log(1e-10)
+
+
+def get_parser():
+    parser = argparse.ArgumentParser(
+        formatter_class=argparse.ArgumentDefaultsHelpFormatter
+    )
+
+    parser.add_argument(
+        "--epoch",
+        type=int,
+        default=28,
+        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=15,
+        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="The experiment dir",
+    )
+
+    parser.add_argument(
+        "--lang-dir",
+        type=str,
+        default="data/lang_char",
+        help="Path to the lang dir(containing lexicon, tokens, etc.)",
+    )
+
+    parser.add_argument(
+        "--decoding-method",
+        type=str,
+        default="greedy_search",
+        help="""Supported decoding methods are:
+        greedy_search
+        modified_beam_search
+        fast_beam_search
+        """,
+    )
+
+    parser.add_argument(
+        "--num_active_paths",
+        type=int,
+        default=4,
+        help="""An interger indicating how many candidates we will keep for each
+        frame. Used only when --decoding-method is 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 --decoding-method is fast_beam_search""",
+    )
+
+    parser.add_argument(
+        "--max-contexts",
+        type=int,
+        default=4,
+        help="""Used only when --decoding-method is
+        fast_beam_search""",
+    )
+
+    parser.add_argument(
+        "--max-states",
+        type=int,
+        default=32,
+        help="""Used only when --decoding-method is
+        fast_beam_search""",
+    )
+
+    parser.add_argument(
+        "--context-size",
+        type=int,
+        default=1,
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
+    )
+
+    parser.add_argument(
+        "--blank-penalty",
+        type=float,
+        default=0.0,
+        help="""
+        The penalty applied on blank symbol during decoding.
+        Note: It is a positive value that would be applied to logits like
+        this `logits[:, 0] -= blank_penalty` (suppose logits.shape is
+        [batch_size, vocab] and blank id is 0).
+        """,
+    )
+
+    parser.add_argument(
+        "--num-decode-streams",
+        type=int,
+        default=2000,
+        help="The number of streams that can be decoded parallel.",
+    )
+
+    add_model_arguments(parser)
+
+    return parser
+
+
+def get_init_states(
+    model: nn.Module,
+    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 = model.encoder.get_init_states(batch_size, device)
+
+    embed_states = model.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
+
+
+def stack_states(state_list: List[List[torch.Tensor]]) -> List[torch.Tensor]:
+    """Stack list of zipformer states that correspond to separate utterances
+    into a single emformer state, so that it can be used as an input for
+    zipformer when those utterances are formed into a batch.
+
+    Args:
+      state_list:
+        Each element in state_list corresponding to the internal state
+        of the zipformer model for a single utterance. For element-n,
+        state_list[n] is a list of cached tensors of all encoder layers. For layer-i,
+        state_list[n][i*6:(i+1)*6] is (cached_key, cached_nonlin_attn, cached_val1,
+        cached_val2, cached_conv1, cached_conv2).
+        state_list[n][-2] is the cached left padding for ConvNeXt module,
+          of shape (batch_size, num_channels, left_pad, num_freqs)
+        state_list[n][-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.
+
+    Note:
+      It is the inverse of :func:`unstack_states`.
+    """
+    batch_size = len(state_list)
+    assert (len(state_list[0]) - 2) % 6 == 0, len(state_list[0])
+    tot_num_layers = (len(state_list[0]) - 2) // 6
+
+    batch_states = []
+    for layer in range(tot_num_layers):
+        layer_offset = layer * 6
+        # cached_key: (left_context_len, batch_size, key_dim)
+        cached_key = torch.cat(
+            [state_list[i][layer_offset] for i in range(batch_size)], dim=1
+        )
+        # cached_nonlin_attn: (num_heads, batch_size, left_context_len, head_dim)
+        cached_nonlin_attn = torch.cat(
+            [state_list[i][layer_offset + 1] for i in range(batch_size)], dim=1
+        )
+        # cached_val1: (left_context_len, batch_size, value_dim)
+        cached_val1 = torch.cat(
+            [state_list[i][layer_offset + 2] for i in range(batch_size)], dim=1
+        )
+        # cached_val2: (left_context_len, batch_size, value_dim)
+        cached_val2 = torch.cat(
+            [state_list[i][layer_offset + 3] for i in range(batch_size)], dim=1
+        )
+        # cached_conv1: (#batch, channels, left_pad)
+        cached_conv1 = torch.cat(
+            [state_list[i][layer_offset + 4] for i in range(batch_size)], dim=0
+        )
+        # cached_conv2: (#batch, channels, left_pad)
+        cached_conv2 = torch.cat(
+            [state_list[i][layer_offset + 5] for i in range(batch_size)], dim=0
+        )
+        batch_states += [
+            cached_key,
+            cached_nonlin_attn,
+            cached_val1,
+            cached_val2,
+            cached_conv1,
+            cached_conv2,
+        ]
+
+    cached_embed_left_pad = torch.cat(
+        [state_list[i][-2] for i in range(batch_size)], dim=0
+    )
+    batch_states.append(cached_embed_left_pad)
+
+    processed_lens = torch.cat([state_list[i][-1] for i in range(batch_size)], dim=0)
+    batch_states.append(processed_lens)
+
+    return batch_states
+
+
+def unstack_states(batch_states: List[Tensor]) -> List[List[Tensor]]:
+    """Unstack the zipformer state corresponding to a batch of utterances
+    into a list of states, where the i-th entry is the state from the i-th
+    utterance in the batch.
+
+    Note:
+      It is the inverse of :func:`stack_states`.
+
+    Args:
+        batch_states: 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).
+          state_list[-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.
+
+    Returns:
+        state_list: A list of list. Each element in state_list corresponding to the internal state
+        of the zipformer model for a single utterance.
+    """
+    assert (len(batch_states) - 2) % 6 == 0, len(batch_states)
+    tot_num_layers = (len(batch_states) - 2) // 6
+
+    processed_lens = batch_states[-1]
+    batch_size = processed_lens.shape[0]
+
+    state_list = [[] for _ in range(batch_size)]
+
+    for layer in range(tot_num_layers):
+        layer_offset = layer * 6
+        # cached_key: (left_context_len, batch_size, key_dim)
+        cached_key_list = batch_states[layer_offset].chunk(chunks=batch_size, dim=1)
+        # cached_nonlin_attn: (num_heads, batch_size, left_context_len, head_dim)
+        cached_nonlin_attn_list = batch_states[layer_offset + 1].chunk(
+            chunks=batch_size, dim=1
+        )
+        # cached_val1: (left_context_len, batch_size, value_dim)
+        cached_val1_list = batch_states[layer_offset + 2].chunk(
+            chunks=batch_size, dim=1
+        )
+        # cached_val2: (left_context_len, batch_size, value_dim)
+        cached_val2_list = batch_states[layer_offset + 3].chunk(
+            chunks=batch_size, dim=1
+        )
+        # cached_conv1: (#batch, channels, left_pad)
+        cached_conv1_list = batch_states[layer_offset + 4].chunk(
+            chunks=batch_size, dim=0
+        )
+        # cached_conv2: (#batch, channels, left_pad)
+        cached_conv2_list = batch_states[layer_offset + 5].chunk(
+            chunks=batch_size, dim=0
+        )
+        for i in range(batch_size):
+            state_list[i] += [
+                cached_key_list[i],
+                cached_nonlin_attn_list[i],
+                cached_val1_list[i],
+                cached_val2_list[i],
+                cached_conv1_list[i],
+                cached_conv2_list[i],
+            ]
+
+    cached_embed_left_pad_list = batch_states[-2].chunk(chunks=batch_size, dim=0)
+    for i in range(batch_size):
+        state_list[i].append(cached_embed_left_pad_list[i])
+
+    processed_lens_list = batch_states[-1].chunk(chunks=batch_size, dim=0)
+    for i in range(batch_size):
+        state_list[i].append(processed_lens_list[i])
+
+    return state_list
+
+
+def streaming_forward(
+    features: Tensor,
+    feature_lens: Tensor,
+    model: nn.Module,
+    states: List[Tensor],
+    chunk_size: int,
+    left_context_len: int,
+) -> Tuple[Tensor, Tensor, List[Tensor]]:
+    """
+    Returns encoder outputs, output lengths, and updated states.
+    """
+    cached_embed_left_pad = states[-2]
+    (x, x_lens, new_cached_embed_left_pad) = model.encoder_embed.streaming_forward(
+        x=features,
+        x_lens=feature_lens,
+        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,
+    ) = model.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
+
+
+def decode_one_chunk(
+    params: AttributeDict,
+    model: nn.Module,
+    decode_streams: List[DecodeStream],
+) -> List[int]:
+    """Decode one chunk frames of features for each decode_streams and
+    return the indexes of finished streams in a List.
+
+    Args:
+      params:
+        It's the return value of :func:`get_params`.
+      model:
+        The neural model.
+      decode_streams:
+        A List of DecodeStream, each belonging to a utterance.
+    Returns:
+      Return a List containing which DecodeStreams are finished.
+    """
+    device = model.device
+    chunk_size = int(params.chunk_size)
+    left_context_len = int(params.left_context_frames)
+
+    features = []
+    feature_lens = []
+    states = []
+    processed_lens = []  # Used in fast-beam-search
+
+    for stream in decode_streams:
+        feat, feat_len = stream.get_feature_frames(chunk_size * 2)
+        features.append(feat)
+        feature_lens.append(feat_len)
+        states.append(stream.states)
+        processed_lens.append(stream.done_frames)
+
+    feature_lens = torch.tensor(feature_lens, device=device)
+    features = pad_sequence(features, batch_first=True, padding_value=LOG_EPS)
+
+    # Make sure the length after encoder_embed is at least 1.
+    # The encoder_embed subsample features (T - 7) // 2
+    # The ConvNeXt module needs (7 - 1) // 2 = 3 frames of right padding after subsampling
+    tail_length = chunk_size * 2 + 7 + 2 * 3
+    if features.size(1) < tail_length:
+        pad_length = tail_length - features.size(1)
+        feature_lens += pad_length
+        features = torch.nn.functional.pad(
+            features,
+            (0, 0, 0, pad_length),
+            mode="constant",
+            value=LOG_EPS,
+        )
+
+    states = stack_states(states)
+
+    encoder_out, encoder_out_lens, new_states = streaming_forward(
+        features=features,
+        feature_lens=feature_lens,
+        model=model,
+        states=states,
+        chunk_size=chunk_size,
+        left_context_len=left_context_len,
+    )
+
+    encoder_out = model.joiner.encoder_proj(encoder_out)
+
+    if params.decoding_method == "greedy_search":
+        greedy_search(
+            model=model,
+            encoder_out=encoder_out,
+            streams=decode_streams,
+            blank_penalty=params.blank_penalty,
+        )
+    elif params.decoding_method == "fast_beam_search":
+        processed_lens = torch.tensor(processed_lens, device=device)
+        processed_lens = processed_lens + encoder_out_lens
+        fast_beam_search_one_best(
+            model=model,
+            encoder_out=encoder_out,
+            processed_lens=processed_lens,
+            streams=decode_streams,
+            beam=params.beam,
+            max_states=params.max_states,
+            max_contexts=params.max_contexts,
+            blank_penalty=params.blank_penalty,
+        )
+    elif params.decoding_method == "modified_beam_search":
+        modified_beam_search(
+            model=model,
+            streams=decode_streams,
+            encoder_out=encoder_out,
+            num_active_paths=params.num_active_paths,
+            blank_penalty=params.blank_penalty,
+        )
+    else:
+        raise ValueError(f"Unsupported decoding method: {params.decoding_method}")
+
+    states = unstack_states(new_states)
+
+    finished_streams = []
+    for i in range(len(decode_streams)):
+        decode_streams[i].states = states[i]
+        decode_streams[i].done_frames += encoder_out_lens[i]
+        if decode_streams[i].done:
+            finished_streams.append(i)
+
+    return finished_streams
+
+
+def decode_dataset(
+    cuts: CutSet,
+    params: AttributeDict,
+    model: nn.Module,
+    lexicon: Lexicon,
+    decoding_graph: Optional[k2.Fsa] = None,
+) -> Dict[str, List[Tuple[List[str], List[str]]]]:
+    """Decode dataset.
+
+    Args:
+      cuts:
+        Lhotse Cutset containing the dataset to decode.
+      params:
+        It is returned by :func:`get_params`.
+      model:
+        The neural model.
+      lexicon:
+        The Lexicon.
+      decoding_graph:
+        The decoding graph. Can be either a `k2.trivial_graph` or HLG, Used
+        only when --decoding_method is fast_beam_search.
+    Returns:
+      Return a dict, whose key may be "greedy_search" if greedy search
+      is used, or it may be "beam_7" if beam size of 7 is used.
+      Its value is a list of tuples. Each tuple contains two elements:
+      The first is the reference transcript, and the second is the
+      predicted result.
+    """
+    device = model.device
+
+    opts = FbankOptions()
+    opts.device = device
+    opts.frame_opts.dither = 0
+    opts.frame_opts.snip_edges = False
+    opts.frame_opts.samp_freq = 16000
+    opts.mel_opts.num_bins = 80
+    opts.mel_opts.high_freq = -400
+
+    log_interval = 100
+
+    decode_results = []
+    # Contain decode streams currently running.
+    decode_streams = []
+    for num, cut in enumerate(cuts):
+        # each utterance has a DecodeStream.
+        initial_states = get_init_states(model=model, batch_size=1, device=device)
+        decode_stream = DecodeStream(
+            params=params,
+            cut_id=cut.id,
+            initial_states=initial_states,
+            decoding_graph=decoding_graph,
+            device=device,
+        )
+
+        audio: np.ndarray = cut.load_audio()
+        # audio.shape: (1, num_samples)
+        assert len(audio.shape) == 2
+        assert audio.shape[0] == 1, "Should be single channel"
+        assert audio.dtype == np.float32, audio.dtype
+
+        # The trained model is using normalized samples
+        # - this is to avoid sending [-32k,+32k] signal in...
+        # - some lhotse AudioTransform classes can make the signal
+        #   be out of range [-1, 1], hence the tolerance 10
+        assert (
+            np.abs(audio).max() <= 10
+        ), "Should be normalized to [-1, 1], 10 for tolerance..."
+
+        samples = torch.from_numpy(audio).squeeze(0)
+
+        fbank = Fbank(opts)
+        feature = fbank(samples.to(device))
+        decode_stream.set_features(feature, tail_pad_len=30)
+        decode_stream.ground_truth = cut.supervisions[0].text
+
+        decode_streams.append(decode_stream)
+
+        while len(decode_streams) >= params.num_decode_streams:
+            finished_streams = decode_one_chunk(
+                params=params, model=model, decode_streams=decode_streams
+            )
+            for i in sorted(finished_streams, reverse=True):
+                decode_results.append(
+                    (
+                        decode_streams[i].id,
+                        list(decode_streams[i].ground_truth.strip()),
+                        [
+                            lexicon.token_table[idx]
+                            for idx in decode_streams[i].decoding_result()
+                        ],
+                    )
+                )
+                del decode_streams[i]
+
+        if num % log_interval == 0:
+            logging.info(f"Cuts processed until now is {num}.")
+
+    # decode final chunks of last sequences
+    while len(decode_streams):
+        finished_streams = decode_one_chunk(
+            params=params, model=model, decode_streams=decode_streams
+        )
+        for i in sorted(finished_streams, reverse=True):
+            decode_results.append(
+                (
+                    decode_streams[i].id,
+                    decode_streams[i].ground_truth.split(),
+                    [
+                        lexicon.token_table[idx]
+                        for idx in decode_streams[i].decoding_result()
+                    ],
+                )
+            )
+            del decode_streams[i]
+
+    key = f"blank_penalty_{params.blank_penalty}"
+    if params.decoding_method == "greedy_search":
+        key = f"greedy_search_{key}"
+    elif params.decoding_method == "fast_beam_search":
+        key = (
+            f"beam_{params.beam}_"
+            f"max_contexts_{params.max_contexts}_"
+            f"max_states_{params.max_states}_{key}"
+        )
+    elif params.decoding_method == "modified_beam_search":
+        key = f"num_active_paths_{params.num_active_paths}_{key}"
+    else:
+        raise ValueError(f"Unsupported decoding method: {params.decoding_method}")
+    return {key: decode_results}
+
+
+def save_results(
+    params: AttributeDict,
+    test_set_name: str,
+    results_dict: Dict[str, List[Tuple[List[str], List[str]]]],
+):
+    test_set_wers = dict()
+    for key, results in results_dict.items():
+        recog_path = (
+            params.res_dir / f"recogs-{test_set_name}-{key}-{params.suffix}.txt"
+        )
+        results = sorted(results)
+        store_transcripts(filename=recog_path, texts=results)
+        logging.info(f"The transcripts are stored in {recog_path}")
+
+        # The following prints out WERs, per-word error statistics and aligned
+        # ref/hyp pairs.
+        errs_filename = (
+            params.res_dir / f"errs-{test_set_name}-{key}-{params.suffix}.txt"
+        )
+        with open(errs_filename, "w") as f:
+            wer = write_error_stats(
+                f, f"{test_set_name}-{key}", results, enable_log=True
+            )
+            test_set_wers[key] = wer
+
+        logging.info("Wrote detailed error stats to {}".format(errs_filename))
+
+    test_set_wers = sorted(test_set_wers.items(), key=lambda x: x[1])
+    errs_info = (
+        params.res_dir / f"wer-summary-{test_set_name}-{key}-{params.suffix}.txt"
+    )
+    with open(errs_info, "w") as f:
+        print("settings\tWER", file=f)
+        for key, val in test_set_wers:
+            print("{}\t{}".format(key, val), file=f)
+
+    s = "\nFor {}, WER of different settings are:\n".format(test_set_name)
+    note = "\tbest for {}".format(test_set_name)
+    for key, val in test_set_wers:
+        s += "{}\t{}{}\n".format(key, val, note)
+        note = ""
+    logging.info(s)
+
+
+@torch.no_grad()
+def main():
+    parser = get_parser()
+    MdccAsrDataModule.add_arguments(parser)
+    args = parser.parse_args()
+    args.exp_dir = Path(args.exp_dir)
+
+    params = get_params()
+    params.update(vars(args))
+
+    params.res_dir = params.exp_dir / "streaming" / params.decoding_method
+
+    if params.iter > 0:
+        params.suffix = f"iter-{params.iter}-avg-{params.avg}"
+    else:
+        params.suffix = f"epoch-{params.epoch}-avg-{params.avg}"
+
+    assert params.causal, 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."
+    params.suffix += f"-chunk-{params.chunk_size}"
+    params.suffix += f"-left-context-{params.left_context_frames}"
+    params.suffix += f"-blank-penalty-{params.blank_penalty}"
+
+    # for fast_beam_search
+    if params.decoding_method == "fast_beam_search":
+        params.suffix += f"-beam-{params.beam}"
+        params.suffix += f"-max-contexts-{params.max_contexts}"
+        params.suffix += f"-max-states-{params.max_states}"
+
+    if params.use_averaged_model:
+        params.suffix += "-use-averaged-model"
+
+    setup_logger(f"{params.res_dir}/log-decode-{params.suffix}")
+    logging.info("Decoding started")
+
+    device = torch.device("cpu")
+    if torch.cuda.is_available():
+        device = torch.device("cuda", 0)
+
+    logging.info(f"Device: {device}")
+
+    lexicon = Lexicon(params.lang_dir)
+    params.blank_id = lexicon.token_table["<blk>"]
+    params.vocab_size = max(lexicon.tokens) + 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.to(device)
+            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 start >= 0:
+                    filenames.append(f"{params.exp_dir}/epoch-{i}.pt")
+            logging.info(f"averaging {filenames}")
+            model.to(device)
+            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.to(device)
+            model.load_state_dict(
+                average_checkpoints_with_averaged_model(
+                    filename_start=filename_start,
+                    filename_end=filename_end,
+                    device=device,
+                )
+            )
+        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.to(device)
+            model.load_state_dict(
+                average_checkpoints_with_averaged_model(
+                    filename_start=filename_start,
+                    filename_end=filename_end,
+                    device=device,
+                )
+            )
+
+    model.to(device)
+    model.eval()
+    model.device = device
+
+    decoding_graph = None
+    if params.decoding_method == "fast_beam_search":
+        decoding_graph = k2.trivial_graph(params.vocab_size - 1, device=device)
+
+    num_param = sum([p.numel() for p in model.parameters()])
+    logging.info(f"Number of model parameters: {num_param}")
+
+    mdcc = MdccAsrDataModule(args)
+
+    valid_cuts = mdcc.valid_cuts()
+    test_cuts = mdcc.test_cuts()
+
+    test_sets = ["valid", "test"]
+    test_cuts = [valid_cuts, test_cuts]
+
+    for test_set, test_cut in zip(test_sets, test_cuts):
+        results_dict = decode_dataset(
+            cuts=test_cut,
+            params=params,
+            model=model,
+            lexicon=lexicon,
+            decoding_graph=decoding_graph,
+        )
+        save_results(
+            params=params,
+            test_set_name=test_set,
+            results_dict=results_dict,
+        )
+
+    logging.info("Done!")
+
+
+if __name__ == "__main__":
+    main()

egs/mdcc/ASR/zipformer/subsampling.py

@@ -0,0 +1 @@
+../../../librispeech/ASR/zipformer/subsampling.py

egs/mdcc/ASR/zipformer/zipformer.py

@@ -0,0 +1 @@
+../../../librispeech/ASR/zipformer/zipformer.py

requirements.txt

@@ -1,4 +1,4 @@
-kaldifst
+kaldifst>1.7.0
 kaldilm
 kaldialign
 num2words
@@ -14,4 +14,7 @@ onnxruntime==1.16.3
 # style check session:
 black==22.3.0
 isort==5.10.1
-flake8==5.0.4 
+flake8==5.0.4 
+
+# cantonese word segment support
+pycantonese==3.4.0