Minor fixes for aishell (#218)

* Minor fixes to aishell.

* Minor fixes.

egs/aishell/ASR/RESULTS.md

@@ -1,49 +1,49 @@
 ## Results
-### Aishell training result(Transducer-stateless) 
+### Aishell training result(Transducer-stateless)
 #### 2022-2-19
-(Duo Ma): The tensorboard log for training is available at https://tensorboard.dev/experiment/25PmX3MxSVGTdvIdhOwllw/#scalars  
+(Duo Ma): The tensorboard log for training is available at https://tensorboard.dev/experiment/25PmX3MxSVGTdvIdhOwllw/#scalars
 You can find a pretrained model by visiting https://huggingface.co/shuanguanma/icefall_aishell_transducer_stateless_context_size2_epoch60_2022_2_19
 |                           | test |comment                                  |
 |---------------------------|------|-----------------------------------------|
 | greedy search             | 5.4 |--epoch 59, --avg 10, --max-duration 100 |
-| beam search               | 5.05|--epoch 59, --avg 10, --max-duration 100 | 
+| beam search               | 5.05|--epoch 59, --avg 10, --max-duration 100 |
 
 You can use the following commands to reproduce our results:
+
+```bash
+export CUDA_VISIBLE_DEVICES="0,1,2,3"
+python3 ./transducer_stateless/train.py \
+      --world-size 4 \
+      --num-epochs 60 \
+      --start-epoch 0 \
+      --exp-dir exp/transducer_stateless_context_size2 \
+      --max-duration 100 \
+      --lr-factor 2.5 \
+      --context-size 2
+
+lang_dir=data/lang_char
+dir=exp/transducer_stateless_context_size2
+python3 ./transducer_stateless/decode.py\
+       --epoch 59 \
+       --avg 10 \
+       --exp-dir $dir \
+       --lang-dir $lang_dir \
+       --decoding-method greedy_search \
+       --context-size 2 \
+       --max-sym-per-frame 3
+
+lang_dir=data/lang_char
+dir=exp/transducer_stateless_context_size2
+python3 ./transducer_stateless/decode.py \
+       --epoch 59\
+       --avg 10\
+       --exp-dir $dir \
+       --lang-dir $lang_dir \
+       --decoding-method beam_search \
+       --context-size 2 \
+       --max-sym-per-frame 3
 ```
-    export CUDA_VISIBLE_DEVICES="0,1,2,3"
-    python3 ./transducer_stateless/train.py \
-          --world-size 4 \
-          --num-epochs 60 \
-          --start-epoch 0 \
-          --exp-dir exp/transducer_stateless_context_size2 \
-          --max-duration 100 \
-          --lr-factor 2.5\
-          --context-size 2
-          
-    lang_dir=data/lang_char
-    dir=exp/transducer_stateless_context_size2
-    python3 ./transducer_stateless/decode.py\
-           --epoch 59\
-           --avg 10\
-           --exp-dir $dir \
-           --lang-dir $lang_dir\
-           --decoding-method greedy_search\
-           --context-size 2\
-           --max-sym-per-frame 3
-    lang_dir=data/lang_char
-    dir=exp/transducer_stateless_context_size2
-    python3 ./transducer_stateless/decode.py\
-           --epoch 59\
-           --avg 10\
-           --exp-dir $dir \
-           --lang-dir $lang_dir\
-           --decoding-method beam_search\
-           --context-size 2\
-           --max-sym-per-frame 3
-  ```
-  
-                
-                
+
 ### Aishell training results (Transducer-stateless)
 #### 2022-02-18
 (Pingfeng Luo) : The tensorboard log for training is available at <https://tensorboard.dev/experiment/SG1KV62hRzO5YZswwMQnoQ/>

egs/aishell/ASR/local/display_manifest_statistics.py

@@ -0,0 +1,118 @@
+#!/usr/bin/env python3
+# Copyright    2021  Xiaomi Corp.        (authors: Fangjun Kuang)
+#
+# See ../../../../LICENSE for clarification regarding multiple authors
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""
+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_stateless/train.py
+for usage.
+"""
+
+
+from lhotse import load_manifest
+
+
+def main():
+    #  path = "./data/fbank/cuts_train.json.gz"
+    #  path = "./data/fbank/cuts_test.json.gz"
+    path = "./data/fbank/cuts_dev.json.gz"
+
+    cuts = load_manifest(path)
+    cuts.describe()
+
+
+if __name__ == "__main__":
+    main()
+
+"""
+## train (after speed perturb)
+Cuts count: 360294
+Total duration (hours): 455.6
+Speech duration (hours): 455.6 (100.0%)
+***
+Duration statistics (seconds):
+mean    4.6
+std     1.4
+min     1.1
+0.1%    1.8
+0.5%    2.2
+1%      2.3
+5%      2.7
+10%     3.0
+10%     3.0
+25%     3.5
+50%     4.3
+75%     5.4
+90%     6.5
+95%     7.2
+99%     8.8
+99.5%   9.4
+99.9%   10.9
+max     16.1
+
+## test
+Cuts count: 7176
+Total duration (hours): 10.0
+Speech duration (hours): 10.0 (100.0%)
+***
+Duration statistics (seconds):
+mean    5.0
+std     1.6
+min     1.9
+0.1%    2.2
+0.5%    2.4
+1%      2.6
+5%      3.0
+10%     3.2
+10%     3.2
+25%     3.8
+50%     4.7
+75%     5.9
+90%     7.3
+95%     8.2
+99%     9.9
+99.5%   10.7
+99.9%   11.9
+max     14.7
+
+## dev
+Cuts count: 14326
+Total duration (hours): 18.1
+Speech duration (hours): 18.1 (100.0%)
+***
+Duration statistics (seconds):
+mean    4.5
+std     1.3
+min     1.6
+0.1%    2.1
+0.5%    2.3
+1%      2.4
+5%      2.9
+10%     3.1
+10%     3.1
+25%     3.5
+50%     4.3
+75%     5.4
+90%     6.4
+95%     7.0
+99%     8.4
+99.5%   8.9
+99.9%   10.3
+max     12.5
+"""

egs/aishell/ASR/prepare.sh

@@ -48,8 +48,9 @@ if [ $stage -le -1 ] && [ $stop_stage -ge -1 ]; then
   log "stage -1: Download LM"
   # We assume that you have installed the git-lfs, if not, you could install it
   # using: `sudo apt-get install git-lfs && git-lfs install`
-  [ ! -e $dl_dir/lm ] && mkdir -p $dl_dir/lm
-  git clone https://huggingface.co/pkufool/aishell_lm $dl_dir/lm
+  if [ ! -f $dl_dir/lm/3-gram.unpruned.arpa ]; then
+    git clone https://huggingface.co/pkufool/aishell_lm $dl_dir/lm
+  fi
 fi
 
 if [ $stage -le 0 ] && [ $stop_stage -ge 0 ]; then
@@ -87,28 +88,41 @@ if [ $stage -le 1 ] && [ $stop_stage -ge 1 ]; then
   log "Stage 1: Prepare aishell manifest"
   # We assume that you have downloaded the aishell corpus
   # to $dl_dir/aishell
-  mkdir -p data/manifests
-  lhotse prepare aishell -j $nj $dl_dir/aishell data/manifests
+  if [ ! -f data/manifests/.aishell_manifests.done ]; then
+    mkdir -p data/manifests
+    lhotse prepare aishell $dl_dir/aishell data/manifests
+    touch data/manifests/.aishell_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
-  mkdir -p data/manifests
-  lhotse prepare musan $dl_dir/musan data/manifests
+  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 aishell"
-  mkdir -p data/fbank
-  ./local/compute_fbank_aishell.py
+  if [ ! -f data/fbank/.aishell.done ]; then
+    mkdir -p data/fbank
+    ./local/compute_fbank_aishell.py
+    touch data/fbank/.aishell.done
+  fi
 fi
 
 if [ $stage -le 4 ] && [ $stop_stage -ge 4 ]; then
   log "Stage 4: Compute fbank for musan"
-  mkdir -p data/fbank
-  ./local/compute_fbank_musan.py
+  if [ ! -f data/fbank/.msuan.done ]; then
+    mkdir -p data/fbank
+    ./local/compute_fbank_musan.py
+    touch data/fbank/.msuan.done
+  fi
 fi
 
 lang_phone_dir=data/lang_phone
@@ -134,7 +148,7 @@ if [ $stage -le 5 ] && [ $stop_stage -ge 5 ]; then
     aishell_train_uid=$dl_dir/aishell/data_aishell/transcript/aishell_train_uid
     find $dl_dir/aishell/data_aishell/wav/train -name "*.wav" | sed 's/\.wav//g' | awk -F '/' '{print $NF}' > $aishell_train_uid
     awk 'NR==FNR{uid[$1]=$1} NR!=FNR{if($1 in uid) print $0}' $aishell_train_uid $aishell_text |
-	cut -d " " -f 2- > $lang_phone_dir/transcript_words.txt
+	    cut -d " " -f 2- > $lang_phone_dir/transcript_words.txt
   fi
 
   if [ ! -f $lang_phone_dir/transcript_tokens.txt ]; then

egs/aishell/ASR/transducer_stateless/train.py

@@ -558,7 +558,6 @@ def run(rank, world_size, args):
         oov="<unk>",
     )
 
-    # params.blank_id = graph_compiler.texts_to_ids("<blk>")[0][0]
     params.blank_id = 0
     params.vocab_size = max(lexicon.tokens) + 1
 

icefall/char_graph_compiler.py

@@ -36,7 +36,7 @@ class CharCtcTrainingGraphCompiler(object):
         """
         Args:
           lexicon:
-            It is built from `data/lang/lexicon.txt`.
+            It is built from `data/lang_char/lexicon.txt`.
           device:
             The device to use for operations compiling transcripts to FSAs.
           oov: