From f293d4ade3eb21c29109dda0ab988866b324ee9c Mon Sep 17 00:00:00 2001
From: Guanbo Wang <wgb14@outlook.com>
Date: Thu, 12 May 2022 00:57:02 -0400
Subject: [PATCH] Update results

---
 egs/gigaspeech/ASR/README.md                  |  7 +-
 egs/gigaspeech/ASR/RESULTS.md                 | 66 +++++++++++++++++++
 .../pruned_transducer_stateless2/decode.py    |  4 +-
 .../ASR/pruned_transducer_stateless2/train.py |  2 +-
 4 files changed, 73 insertions(+), 6 deletions(-)

diff --git a/egs/gigaspeech/ASR/README.md b/egs/gigaspeech/ASR/README.md
index 7796ef2a0..1fca69e8b 100644
--- a/egs/gigaspeech/ASR/README.md
+++ b/egs/gigaspeech/ASR/README.md
@@ -13,8 +13,9 @@ ln -sfv /path/to/GigaSpeech download/GigaSpeech
 ```
 
 ## Performance Record
-|     |  Dev  | Test  |
-|-----|-------|-------|
-| WER | 10.47 | 10.58 |
+|                                |  Dev  | Test  |
+|--------------------------------|-------|-------|
+|         `conformer_ctc`        | 10.47 | 10.58 |
+| `pruned_transducer_stateless2` | 10.52 | 10.62 |
 
 See [RESULTS](/egs/gigaspeech/ASR/RESULTS.md) for details.
diff --git a/egs/gigaspeech/ASR/RESULTS.md b/egs/gigaspeech/ASR/RESULTS.md
index b29e893da..6026eddc5 100644
--- a/egs/gigaspeech/ASR/RESULTS.md
+++ b/egs/gigaspeech/ASR/RESULTS.md
@@ -1,4 +1,70 @@
 ## Results
+### GigaSpeech BPE training results (Pruned Transducer 2)
+
+#### 2022-05-12
+
+Results are:
+
+|                      |  Dev  | Test  |
+|----------------------|-------|-------|
+|    greedy search     | 10.59 | 10.87 |
+|   fast beam search   | 10.56 | 10.80 |
+| modified beam search | 10.52 | 10.62 |
+
+To reproduce the above result, use the following commands for training:
+
+```
+cd egs/gigaspeech/ASR
+./prepare.sh
+export CUDA_VISIBLE_DEVICES="0,1,2,3,4,5,6,7"
+./pruned_transducer_stateless2/train.py \
+  --max-duration 120 \
+  --num-workers 1 \
+  --world-size 8 \
+  --exp-dir pruned_transducer_stateless2/exp \
+  --bpe-model data/lang_bpe_500/bpe.model \
+  --use-fp16 True
+```
+
+and the following commands for decoding:
+
+```
+# greedy search
+./pruned_transducer_stateless2/decode.py \
+  --epoch 29 \
+  --avg 11 \
+  --decoding-method greedy_search \
+  --exp-dir pruned_transducer_stateless2/exp \
+  --bpe-model data/lang_bpe_500/bpe.model \
+  --max-duration 20 \
+  --num-workers 1
+
+# fast beam search
+./pruned_transducer_stateless2/decode.py \
+  --epoch 29 \
+  --avg 9 \
+  --decoding-method fast_beam_search \
+  --exp-dir pruned_transducer_stateless2/exp \
+  --bpe-model data/lang_bpe_500/bpe.model \
+  --max-duration 20 \
+  --num-workers 1
+
+# modified beam search
+./pruned_transducer_stateless2/decode.py \
+  --epoch 29 \
+  --avg 8 \
+  --decoding-method modified_beam_search \
+  --exp-dir pruned_transducer_stateless2/exp \
+  --bpe-model data/lang_bpe_500/bpe.model \
+  --max-duration 20 \
+  --num-workers 1
+```
+
+Pretrained model is available at
+<https://huggingface.co/wgb14/icefall-asr-gigaspeech-pruned-transducer-stateless2>
+
+The tensorboard log for training is available at
+<https://tensorboard.dev/experiment/zmmM0MLASnG1N2RmJ4MZBw/>
 
 ### GigaSpeech BPE training results (Conformer-CTC)
 
diff --git a/egs/gigaspeech/ASR/pruned_transducer_stateless2/decode.py b/egs/gigaspeech/ASR/pruned_transducer_stateless2/decode.py
index f36cc95f7..076feb7a9 100755
--- a/egs/gigaspeech/ASR/pruned_transducer_stateless2/decode.py
+++ b/egs/gigaspeech/ASR/pruned_transducer_stateless2/decode.py
@@ -98,14 +98,14 @@ def get_parser():
     parser.add_argument(
         "--epoch",
         type=int,
-        default=28,
+        default=29,
         help="It specifies the checkpoint to use for decoding."
         "Note: Epoch counts from 0.",
     )
     parser.add_argument(
         "--avg",
         type=int,
-        default=15,
+        default=8,
         help="Number of checkpoints to average. Automatically select "
         "consecutive checkpoints before the checkpoint specified by "
         "'--epoch'. ",
diff --git a/egs/gigaspeech/ASR/pruned_transducer_stateless2/train.py b/egs/gigaspeech/ASR/pruned_transducer_stateless2/train.py
index 443222683..4421ce2aa 100755
--- a/egs/gigaspeech/ASR/pruned_transducer_stateless2/train.py
+++ b/egs/gigaspeech/ASR/pruned_transducer_stateless2/train.py
@@ -111,7 +111,7 @@ def get_parser():
     parser.add_argument(
         "--num-epochs",
         type=int,
-        default=20,
+        default=30,
         help="Number of epochs to train.",
     )