test

.flake8

@@ -24,6 +24,7 @@ exclude =
   **/data/**,
   icefall/shared/make_kn_lm.py,
   icefall/__init__.py
+  icefall/ctc/__init__.py
 
 ignore =
   # E203 white space before ":"

.github/scripts/run-gigaspeech-pruned-transducer-stateless2-2022-05-12.sh

@@ -29,6 +29,9 @@ if [[ x"${GITHUB_EVENT_NAME}" == x"schedule" || x"${GITHUB_EVENT_LABEL_NAME}" ==
   ls -lh data/fbank
   ls -lh pruned_transducer_stateless2/exp
 
+  ln -s data/fbank/cuts_DEV.jsonl.gz data/fbank/gigaspeech_cuts_DEV.jsonl.gz
+  ln -s data/fbank/cuts_TEST.jsonl.gz data/fbank/gigaspeech_cuts_TEST.jsonl.gz
+
   log "Decoding dev and test"
 
   # use a small value for decoding with CPU

.github/scripts/run-librispeech-conformer-ctc3-2022-11-28.sh

@@ -38,7 +38,7 @@ log "Decode with models exported by torch.jit.trace()"
 for m in ctc-decoding 1best; do
   ./conformer_ctc3/jit_pretrained.py \
     --model-filename $repo/exp/jit_trace.pt \
-    --words-file $repo/data/lang_bpe_500/words.txt  \
+    --words-file $repo/data/lang_bpe_500/words.txt \
     --HLG $repo/data/lang_bpe_500/HLG.pt \
     --bpe-model $repo/data/lang_bpe_500/bpe.model \
     --G $repo/data/lm/G_4_gram.pt \
@@ -53,7 +53,7 @@ log "Export to torchscript model"
 
 ./conformer_ctc3/export.py \
   --exp-dir $repo/exp \
-  --lang-dir $repo/data/lang_bpe_500 \
+  --tokens $repo/data/lang_bpe_500/tokens.txt \
   --jit-trace 1 \
   --epoch 99 \
   --avg 1 \
@@ -80,9 +80,9 @@ done
 for m in ctc-decoding 1best; do
   ./conformer_ctc3/pretrained.py \
     --checkpoint $repo/exp/pretrained.pt \
-    --words-file $repo/data/lang_bpe_500/words.txt  \
+    --words-file $repo/data/lang_bpe_500/words.txt \
     --HLG $repo/data/lang_bpe_500/HLG.pt \
-    --bpe-model $repo/data/lang_bpe_500/bpe.model \
+    --tokens $repo/data/lang_bpe_500/tokens.txt \
     --G $repo/data/lm/G_4_gram.pt \
     --method $m \
     --sample-rate 16000 \
@@ -93,7 +93,7 @@ done
 
 echo "GITHUB_EVENT_NAME: ${GITHUB_EVENT_NAME}"
 echo "GITHUB_EVENT_LABEL_NAME: ${GITHUB_EVENT_LABEL_NAME}"
-if [[ x"${GITHUB_EVENT_NAME}" == x"schedule" || x"${GITHUB_EVENT_LABEL_NAME}" == x"run-decode"  ]]; then
+if [[ x"${GITHUB_EVENT_NAME}" == x"schedule" || x"${GITHUB_EVENT_LABEL_NAME}" == x"run-decode" ]]; then
   mkdir -p conformer_ctc3/exp
   ln -s $PWD/$repo/exp/pretrained.pt conformer_ctc3/exp/epoch-999.pt
   ln -s $PWD/$repo/data/lang_bpe_500 data/

.github/scripts/run-librispeech-lstm-transducer-stateless2-2022-09-03.sh

@@ -31,7 +31,7 @@ log "Test exporting with torch.jit.trace()"
 
 ./lstm_transducer_stateless2/export.py \
   --exp-dir $repo/exp \
-  --bpe-model $repo/data/lang_bpe_500/bpe.model \
+  --tokens $repo/data/lang_bpe_500/tokens.txt \
   --epoch 99 \
   --avg 1 \
   --use-averaged-model 0 \
@@ -55,7 +55,7 @@ for sym in 1 2 3; do
     --method greedy_search \
     --max-sym-per-frame $sym \
     --checkpoint $repo/exp/pretrained.pt \
-    --bpe-model $repo/data/lang_bpe_500/bpe.model \
+    --tokens $repo/data/lang_bpe_500/tokens.txt \
     $repo/test_wavs/1089-134686-0001.wav \
     $repo/test_wavs/1221-135766-0001.wav \
     $repo/test_wavs/1221-135766-0002.wav
@@ -68,7 +68,7 @@ for method in modified_beam_search beam_search fast_beam_search; do
     --method $method \
     --beam-size 4 \
     --checkpoint $repo/exp/pretrained.pt \
-    --bpe-model $repo/data/lang_bpe_500/bpe.model \
+    --tokens $repo/data/lang_bpe_500/tokens.txt \
     $repo/test_wavs/1089-134686-0001.wav \
     $repo/test_wavs/1221-135766-0001.wav \
     $repo/test_wavs/1221-135766-0002.wav

.github/scripts/run-librispeech-pruned-transducer-stateless-2022-03-12.sh

@@ -28,7 +28,7 @@ for sym in 1 2 3; do
     --method greedy_search \
     --max-sym-per-frame $sym \
     --checkpoint $repo/exp/pretrained.pt \
-    --bpe-model $repo/data/lang_bpe_500/bpe.model \
+    --tokens $repo/data/lang_bpe_500/tokens.txt \
     $repo/test_wavs/1089-134686-0001.wav \
     $repo/test_wavs/1221-135766-0001.wav \
     $repo/test_wavs/1221-135766-0002.wav
@@ -41,7 +41,7 @@ for method in fast_beam_search modified_beam_search beam_search; do
     --method $method \
     --beam-size 4 \
     --checkpoint $repo/exp/pretrained.pt \
-    --bpe-model $repo/data/lang_bpe_500/bpe.model \
+    --tokens $repo/data/lang_bpe_500/tokens.txt \
     $repo/test_wavs/1089-134686-0001.wav \
     $repo/test_wavs/1221-135766-0001.wav \
     $repo/test_wavs/1221-135766-0002.wav

.github/scripts/run-librispeech-pruned-transducer-stateless2-2022-04-29.sh

@@ -36,7 +36,7 @@ for sym in 1 2 3; do
     --method greedy_search \
     --max-sym-per-frame $sym \
     --checkpoint $repo/exp/pretrained.pt \
-    --bpe-model $repo/data/lang_bpe_500/bpe.model \
+    --tokens $repo/data/lang_bpe_500/tokens.txt \
     $repo/test_wavs/1089-134686-0001.wav \
     $repo/test_wavs/1221-135766-0001.wav \
     $repo/test_wavs/1221-135766-0002.wav
@@ -49,7 +49,7 @@ for method in modified_beam_search beam_search fast_beam_search; do
     --method $method \
     --beam-size 4 \
     --checkpoint $repo/exp/pretrained.pt \
-    --bpe-model $repo/data/lang_bpe_500/bpe.model \
+    --tokens $repo/data/lang_bpe_500/tokens.txt \
     $repo/test_wavs/1089-134686-0001.wav \
     $repo/test_wavs/1221-135766-0001.wav \
     $repo/test_wavs/1221-135766-0002.wav

.github/scripts/run-librispeech-pruned-transducer-stateless3-2022-04-29.sh

@@ -35,7 +35,7 @@ for sym in 1 2 3; do
     --method greedy_search \
     --max-sym-per-frame $sym \
     --checkpoint $repo/exp/pretrained.pt \
-    --bpe-model $repo/data/lang_bpe_500/bpe.model \
+    --tokens $repo/data/lang_bpe_500/tokens.txt \
     $repo/test_wavs/1089-134686-0001.wav \
     $repo/test_wavs/1221-135766-0001.wav \
     $repo/test_wavs/1221-135766-0002.wav
@@ -48,7 +48,7 @@ for method in modified_beam_search beam_search fast_beam_search; do
     --method $method \
     --beam-size 4 \
     --checkpoint $repo/exp/pretrained.pt \
-    --bpe-model $repo/data/lang_bpe_500/bpe.model \
+    --tokens $repo/data/lang_bpe_500/tokens.txt \
     $repo/test_wavs/1089-134686-0001.wav \
     $repo/test_wavs/1221-135766-0001.wav \
     $repo/test_wavs/1221-135766-0002.wav

.github/scripts/run-librispeech-pruned-transducer-stateless3-2022-05-13.sh

@@ -30,14 +30,14 @@ popd
 log "Export to torchscript model"
 ./pruned_transducer_stateless3/export.py \
   --exp-dir $repo/exp \
-  --bpe-model $repo/data/lang_bpe_500/bpe.model \
+  --tokens $repo/data/lang_bpe_500/tokens.txt \
   --epoch 99 \
   --avg 1 \
   --jit 1
 
 ./pruned_transducer_stateless3/export.py \
   --exp-dir $repo/exp \
-  --bpe-model $repo/data/lang_bpe_500/bpe.model \
+  --tokens $repo/data/lang_bpe_500/tokens.txt \
   --epoch 99 \
   --avg 1 \
   --jit-trace 1
@@ -74,7 +74,7 @@ for sym in 1 2 3; do
     --method greedy_search \
     --max-sym-per-frame $sym \
     --checkpoint $repo/exp/pretrained.pt \
-    --bpe-model $repo/data/lang_bpe_500/bpe.model \
+    --tokens $repo/data/lang_bpe_500/tokens.txt \
     $repo/test_wavs/1089-134686-0001.wav \
     $repo/test_wavs/1221-135766-0001.wav \
     $repo/test_wavs/1221-135766-0002.wav
@@ -87,7 +87,7 @@ for method in modified_beam_search beam_search fast_beam_search; do
     --method $method \
     --beam-size 4 \
     --checkpoint $repo/exp/pretrained.pt \
-    --bpe-model $repo/data/lang_bpe_500/bpe.model \
+    --tokens $repo/data/lang_bpe_500/tokens.txt \
     $repo/test_wavs/1089-134686-0001.wav \
     $repo/test_wavs/1221-135766-0001.wav \
     $repo/test_wavs/1221-135766-0002.wav

.github/scripts/run-librispeech-pruned-transducer-stateless5-2022-05-13.sh

@@ -32,7 +32,7 @@ for sym in 1 2 3; do
     --method greedy_search \
     --max-sym-per-frame $sym \
     --checkpoint $repo/exp/pretrained.pt \
-    --bpe-model $repo/data/lang_bpe_500/bpe.model \
+    --tokens $repo/data/lang_bpe_500/tokens.txt \
     --num-encoder-layers 18 \
     --dim-feedforward 2048 \
     --nhead 8 \
@@ -51,7 +51,7 @@ for method in modified_beam_search beam_search fast_beam_search; do
     --method $method \
     --beam-size 4 \
     --checkpoint $repo/exp/pretrained.pt \
-    --bpe-model $repo/data/lang_bpe_500/bpe.model \
+    --tokens $repo/data/lang_bpe_500/tokens.txt \
     $repo/test_wavs/1089-134686-0001.wav \
     $repo/test_wavs/1221-135766-0001.wav \
     $repo/test_wavs/1221-135766-0002.wav \

.github/scripts/run-librispeech-pruned-transducer-stateless7-2022-11-11.sh

@@ -33,7 +33,7 @@ log "Export to torchscript model"
 ./pruned_transducer_stateless7/export.py \
   --exp-dir $repo/exp \
   --use-averaged-model false \
-  --bpe-model $repo/data/lang_bpe_500/bpe.model \
+  --tokens $repo/data/lang_bpe_500/tokens.txt \
   --epoch 99 \
   --avg 1 \
   --jit 1
@@ -56,7 +56,7 @@ for sym in 1 2 3; do
     --method greedy_search \
     --max-sym-per-frame $sym \
     --checkpoint $repo/exp/pretrained.pt \
-    --bpe-model $repo/data/lang_bpe_500/bpe.model \
+    --tokens $repo/data/lang_bpe_500/tokens.txt \
     $repo/test_wavs/1089-134686-0001.wav \
     $repo/test_wavs/1221-135766-0001.wav \
     $repo/test_wavs/1221-135766-0002.wav
@@ -69,7 +69,7 @@ for method in modified_beam_search beam_search fast_beam_search; do
     --method $method \
     --beam-size 4 \
     --checkpoint $repo/exp/pretrained.pt \
-    --bpe-model $repo/data/lang_bpe_500/bpe.model \
+    --tokens $repo/data/lang_bpe_500/tokens.txt \
     $repo/test_wavs/1089-134686-0001.wav \
     $repo/test_wavs/1221-135766-0001.wav \
     $repo/test_wavs/1221-135766-0002.wav

.github/scripts/run-librispeech-pruned-transducer-stateless7-ctc-2022-12-01.sh

@@ -37,7 +37,7 @@ log "Export to torchscript model"
 ./pruned_transducer_stateless7_ctc/export.py \
   --exp-dir $repo/exp \
   --use-averaged-model false \
-  --bpe-model $repo/data/lang_bpe_500/bpe.model \
+  --tokens $repo/data/lang_bpe_500/tokens.txt \
   --epoch 99 \
   --avg 1 \
   --jit 1
@@ -74,7 +74,7 @@ for sym in 1 2 3; do
     --method greedy_search \
     --max-sym-per-frame $sym \
     --checkpoint $repo/exp/pretrained.pt \
-    --bpe-model $repo/data/lang_bpe_500/bpe.model \
+    --tokens $repo/data/lang_bpe_500/tokens.txt \
     $repo/test_wavs/1089-134686-0001.wav \
     $repo/test_wavs/1221-135766-0001.wav \
     $repo/test_wavs/1221-135766-0002.wav
@@ -87,7 +87,7 @@ for method in modified_beam_search beam_search fast_beam_search; do
     --method $method \
     --beam-size 4 \
     --checkpoint $repo/exp/pretrained.pt \
-    --bpe-model $repo/data/lang_bpe_500/bpe.model \
+    --tokens $repo/data/lang_bpe_500/tokens.txt \
     $repo/test_wavs/1089-134686-0001.wav \
     $repo/test_wavs/1221-135766-0001.wav \
     $repo/test_wavs/1221-135766-0002.wav

.github/scripts/run-librispeech-pruned-transducer-stateless7-ctc-bs-2022-12-15.sh → .github/scripts/run-librispeech-pruned-transducer-stateless7-ctc-bs-2023-01-29.sh

@@ -21,9 +21,9 @@ tree $repo/
 ls -lh $repo/test_wavs/*.wav
 
 pushd $repo/exp
-git lfs pull --include "data/lang_bpe_500/HLG.pt"
 git lfs pull --include "data/lang_bpe_500/L.pt"
 git lfs pull --include "data/lang_bpe_500/LG.pt"
+git lfs pull --include "data/lang_bpe_500/HLG.pt"
 git lfs pull --include "data/lang_bpe_500/Linv.pt"
 git lfs pull --include "data/lang_bpe_500/bpe.model"
 git lfs pull --include "exp/cpu_jit.pt"
@@ -36,7 +36,7 @@ log "Export to torchscript model"
 ./pruned_transducer_stateless7_ctc_bs/export.py \
   --exp-dir $repo/exp \
   --use-averaged-model false \
-  --bpe-model $repo/data/lang_bpe_500/bpe.model \
+  --tokens $repo/data/lang_bpe_500/tokens.txt \
   --epoch 99 \
   --avg 1 \
   --jit 1
@@ -72,7 +72,7 @@ for sym in 1 2 3; do
     --method greedy_search \
     --max-sym-per-frame $sym \
     --checkpoint $repo/exp/pretrained.pt \
-    --bpe-model $repo/data/lang_bpe_500/bpe.model \
+    --tokens $repo/data/lang_bpe_500/tokens.txt \
     $repo/test_wavs/1089-134686-0001.wav \
     $repo/test_wavs/1221-135766-0001.wav \
     $repo/test_wavs/1221-135766-0002.wav
@@ -85,7 +85,7 @@ for method in modified_beam_search beam_search fast_beam_search; do
     --method $method \
     --beam-size 4 \
     --checkpoint $repo/exp/pretrained.pt \
-    --bpe-model $repo/data/lang_bpe_500/bpe.model \
+    --tokens $repo/data/lang_bpe_500/tokens.txt \
     $repo/test_wavs/1089-134686-0001.wav \
     $repo/test_wavs/1221-135766-0001.wav \
     $repo/test_wavs/1221-135766-0002.wav

.github/scripts/run-librispeech-pruned-transducer-stateless7-streaming-2022-12-29.sh

@@ -37,7 +37,7 @@ log "Export to torchscript model"
 ./pruned_transducer_stateless7_streaming/export.py \
   --exp-dir $repo/exp \
   --use-averaged-model false \
-  --bpe-model $repo/data/lang_bpe_500/bpe.model \
+  --tokens $repo/data/lang_bpe_500/tokens.txt \
   --decode-chunk-len 32 \
   --epoch 99 \
   --avg 1 \
@@ -81,7 +81,7 @@ for sym in 1 2 3; do
     --method greedy_search \
     --max-sym-per-frame $sym \
     --checkpoint $repo/exp/pretrained.pt \
-    --bpe-model $repo/data/lang_bpe_500/bpe.model \
+    --tokens $repo/data/lang_bpe_500/tokens.txt \
     --decode-chunk-len 32 \
     $repo/test_wavs/1089-134686-0001.wav \
     $repo/test_wavs/1221-135766-0001.wav \
@@ -95,7 +95,7 @@ for method in modified_beam_search beam_search fast_beam_search; do
     --method $method \
     --beam-size 4 \
     --checkpoint $repo/exp/pretrained.pt \
-    --bpe-model $repo/data/lang_bpe_500/bpe.model \
+    --tokens $repo/data/lang_bpe_500/tokens.txt \
     --decode-chunk-len 32 \
     $repo/test_wavs/1089-134686-0001.wav \
     $repo/test_wavs/1221-135766-0001.wav \

.github/scripts/run-librispeech-pruned-transducer-stateless8-2022-11-14.sh

@@ -41,7 +41,7 @@ log "Decode with models exported by torch.jit.script()"
 log "Export to torchscript model"
 ./pruned_transducer_stateless8/export.py \
   --exp-dir $repo/exp \
-  --bpe-model $repo/data/lang_bpe_500/bpe.model \
+  --tokens $repo/data/lang_bpe_500/tokens.txt \
   --use-averaged-model false \
   --epoch 99 \
   --avg 1 \
@@ -65,7 +65,7 @@ for sym in 1 2 3; do
     --method greedy_search \
     --max-sym-per-frame $sym \
     --checkpoint $repo/exp/pretrained.pt \
-    --bpe-model $repo/data/lang_bpe_500/bpe.model \
+    --tokens $repo/data/lang_bpe_500/tokens.txt \
     $repo/test_wavs/1089-134686-0001.wav \
     $repo/test_wavs/1221-135766-0001.wav \
     $repo/test_wavs/1221-135766-0002.wav
@@ -78,7 +78,7 @@ for method in modified_beam_search beam_search fast_beam_search; do
     --method $method \
     --beam-size 4 \
     --checkpoint $repo/exp/pretrained.pt \
-    --bpe-model $repo/data/lang_bpe_500/bpe.model \
+    --tokens $repo/data/lang_bpe_500/tokens.txt \
     $repo/test_wavs/1089-134686-0001.wav \
     $repo/test_wavs/1221-135766-0001.wav \
     $repo/test_wavs/1221-135766-0002.wav

.github/scripts/run-librispeech-streaming-pruned-transducer-stateless2-2022-06-26.sh

@@ -32,7 +32,7 @@ for sym in 1 2 3; do
     --method greedy_search \
     --max-sym-per-frame $sym \
     --checkpoint $repo/exp/pretrained.pt \
-    --bpe-model $repo/data/lang_bpe_500/bpe.model \
+    --tokens $repo/data/lang_bpe_500/tokens.txt \
     --simulate-streaming 1 \
     --causal-convolution 1 \
     $repo/test_wavs/1089-134686-0001.wav \
@@ -47,7 +47,7 @@ for method in modified_beam_search beam_search fast_beam_search; do
     --method $method \
     --beam-size 4 \
     --checkpoint $repo/exp/pretrained.pt \
-    --bpe-model $repo/data/lang_bpe_500/bpe.model \
+    --tokens $repo/data/lang_bpe_500/tokens.txt \
     --simulate-streaming 1 \
     --causal-convolution 1 \
     $repo/test_wavs/1089-134686-0001.wav \

.github/scripts/run-librispeech-streaming-zipformer-2023-05-18.sh

@@ -23,6 +23,7 @@ ls -lh $repo/test_wavs/*.wav
 
 pushd $repo/exp
 git lfs pull --include "data/lang_bpe_500/bpe.model"
+git lfs pull --include "data/lang_bpe_500/tokens.txt"
 git lfs pull --include "exp/jit_script_chunk_16_left_128.pt"
 git lfs pull --include "exp/pretrained.pt"
 ln -s pretrained.pt epoch-99.pt
@@ -33,7 +34,7 @@ log "Export to torchscript model"
 ./zipformer/export.py \
   --exp-dir $repo/exp \
   --use-averaged-model false \
-  --bpe-model $repo/data/lang_bpe_500/bpe.model \
+  --tokens $repo/data/lang_bpe_500/tokens.txt \
   --causal 1 \
   --chunk-size 16 \
   --left-context-frames 128 \
@@ -46,7 +47,7 @@ ls -lh $repo/exp/*.pt
 log "Decode with models exported by torch.jit.script()"
 
 ./zipformer/jit_pretrained_streaming.py \
-  --bpe-model $repo/data/lang_bpe_500/bpe.model \
+  --tokens $repo/data/lang_bpe_500/tokens.txt \
   --nn-model-filename $repo/exp/jit_script_chunk_16_left_128.pt \
   $repo/test_wavs/1089-134686-0001.wav
 
@@ -60,7 +61,7 @@ for method in greedy_search modified_beam_search fast_beam_search; do
     --method $method \
     --beam-size 4 \
     --checkpoint $repo/exp/pretrained.pt \
-    --bpe-model $repo/data/lang_bpe_500/bpe.model \
+    --tokens $repo/data/lang_bpe_500/tokens.txt \
     $repo/test_wavs/1089-134686-0001.wav \
     $repo/test_wavs/1221-135766-0001.wav \
     $repo/test_wavs/1221-135766-0002.wav