shuffle full Librispeech data (#574)

* shuffled full/partial librispeech data

* fixed the code style issue

* Shuffled full librispeech data off-line

* Fixed style, addressed comments, and removed redandunt codes

* Used the suggested version of black

* Propagated the changes to other folders for librispeech (except
conformer_mmi and streaming_conformer_ctc)

egs/librispeech/ASR/conformer_ctc/train.py

@@ -687,10 +687,10 @@ def run(rank, world_size, args):
 
     librispeech = LibriSpeechAsrDataModule(args)
 
-    train_cuts = librispeech.train_clean_100_cuts()
     if params.full_libri:
-        train_cuts += librispeech.train_clean_360_cuts()
-        train_cuts += librispeech.train_other_500_cuts()
+        train_cuts = librispeech.train_all_shuf_cuts()
+    else:
+        train_cuts = librispeech.train_clean_100_cuts()
 
     def remove_short_and_long_utt(c: Cut):
         # Keep only utterances with duration between 1 second and 20 seconds

egs/librispeech/ASR/conformer_ctc2/train.py

@@ -928,10 +928,10 @@ def run(rank, world_size, args):
 
     librispeech = LibriSpeechAsrDataModule(args)
 
-    train_cuts = librispeech.train_clean_100_cuts()
     if params.full_libri:
-        train_cuts += librispeech.train_clean_360_cuts()
-        train_cuts += librispeech.train_other_500_cuts()
+        train_cuts = librispeech.train_all_shuf_cuts()
+    else:
+        train_cuts = librispeech.train_clean_100_cuts()
 
     def remove_short_and_long_utt(c: Cut):
         # Keep only utterances with duration between 1 second and 20 seconds

egs/librispeech/ASR/conv_emformer_transducer_stateless/train.py

@@ -970,10 +970,10 @@ def run(rank, world_size, args):
 
     librispeech = LibriSpeechAsrDataModule(args)
 
-    train_cuts = librispeech.train_clean_100_cuts()
     if params.full_libri:
-        train_cuts += librispeech.train_clean_360_cuts()
-        train_cuts += librispeech.train_other_500_cuts()
+        train_cuts = librispeech.train_all_shuf_cuts()
+    else:
+        train_cuts = librispeech.train_clean_100_cuts()
 
     def remove_short_and_long_utt(c: Cut):
         # Keep only utterances with duration between 1 second and 20 seconds

egs/librispeech/ASR/conv_emformer_transducer_stateless2/train.py

@@ -970,10 +970,10 @@ def run(rank, world_size, args):
 
     librispeech = LibriSpeechAsrDataModule(args)
 
-    train_cuts = librispeech.train_clean_100_cuts()
     if params.full_libri:
-        train_cuts += librispeech.train_clean_360_cuts()
-        train_cuts += librispeech.train_other_500_cuts()
+        train_cuts = librispeech.train_all_shuf_cuts()
+    else:
+        train_cuts = librispeech.train_clean_100_cuts()
 
     def remove_short_and_long_utt(c: Cut):
         # Keep only utterances with duration between 1 second and 20 seconds

egs/librispeech/ASR/lstm_transducer_stateless/train.py

@@ -954,10 +954,10 @@ def run(rank, world_size, args):
 
     librispeech = LibriSpeechAsrDataModule(args)
 
-    train_cuts = librispeech.train_clean_100_cuts()
     if params.full_libri:
-        train_cuts += librispeech.train_clean_360_cuts()
-        train_cuts += librispeech.train_other_500_cuts()
+        train_cuts = librispeech.train_all_shuf_cuts()
+    else:
+        train_cuts = librispeech.train_clean_100_cuts()
 
     def remove_short_and_long_utt(c: Cut):
         # Keep only utterances with duration between 1 second and 20 seconds

egs/librispeech/ASR/lstm_transducer_stateless2/train.py

@@ -1108,10 +1108,10 @@ def run(rank, world_size, args):
 
     librispeech = LibriSpeech(manifest_dir=args.manifest_dir)
 
-    train_cuts = librispeech.train_clean_100_cuts()
     if params.full_libri:
-        train_cuts += librispeech.train_clean_360_cuts()
-        train_cuts += librispeech.train_other_500_cuts()
+        train_cuts = librispeech.train_all_shuf_cuts()
+    else:
+        train_cuts = librispeech.train_clean_100_cuts()
 
     train_cuts = filter_short_and_long_utterances(train_cuts, sp)
 

egs/librispeech/ASR/prepare.sh

@@ -123,6 +123,11 @@ if [ $stage -le 3 ] && [ $stop_stage -ge 3 ]; then
     touch data/fbank/.librispeech.done
   fi
 
+  cat <(gunzip -c data/fbank/librispeech_cuts_train-clean-100.jsonl.gz) \
+    <(gunzip -c data/fbank/librispeech_cuts_train-clean-360.jsonl.gz) \
+    <(gunzip -c data/fbank/librispeech_cuts_train-other-500.jsonl.gz) | \
+    shuf | gzip -c > data/fbank/librispeech_cuts_train-all-shuf.jsonl.gz
+
   if [ ! -e data/fbank/.librispeech-validated.done ]; then
     log "Validating data/fbank for LibriSpeech"
     parts=(

egs/librispeech/ASR/pruned_stateless_emformer_rnnt2/train.py

@@ -882,10 +882,10 @@ def run(rank, world_size, args):
 
     librispeech = LibriSpeechAsrDataModule(args)
 
-    train_cuts = librispeech.train_clean_100_cuts()
     if params.full_libri:
-        train_cuts += librispeech.train_clean_360_cuts()
-        train_cuts += librispeech.train_other_500_cuts()
+        train_cuts = librispeech.train_all_shuf_cuts()
+    else:
+        train_cuts = librispeech.train_clean_100_cuts()
 
     def remove_short_and_long_utt(c: Cut):
         # Keep only utterances with duration between 1 second and 20 seconds

egs/librispeech/ASR/pruned_transducer_stateless/train.py

@@ -873,10 +873,10 @@ def run(rank, world_size, args):
 
     librispeech = LibriSpeechAsrDataModule(args)
 
-    train_cuts = librispeech.train_clean_100_cuts()
     if params.full_libri:
-        train_cuts += librispeech.train_clean_360_cuts()
-        train_cuts += librispeech.train_other_500_cuts()
+        train_cuts = librispeech.train_all_shuf_cuts()
+    else:
+        train_cuts = librispeech.train_clean_100_cuts()
 
     def remove_short_and_long_utt(c: Cut):
         # Keep only utterances with duration between 1 second and 20 seconds

egs/librispeech/ASR/pruned_transducer_stateless2/train.py

@@ -931,10 +931,10 @@ def run(rank, world_size, args):
 
     librispeech = LibriSpeechAsrDataModule(args)
 
-    train_cuts = librispeech.train_clean_100_cuts()
     if params.full_libri:
-        train_cuts += librispeech.train_clean_360_cuts()
-        train_cuts += librispeech.train_other_500_cuts()
+        train_cuts = librispeech.train_all_shuf_cuts()
+    else:
+        train_cuts = librispeech.train_clean_100_cuts()
 
     def remove_short_and_long_utt(c: Cut):
         # Keep only utterances with duration between 1 second and 20 seconds

egs/librispeech/ASR/pruned_transducer_stateless3/train.py

@@ -1065,10 +1065,10 @@ def run(rank, world_size, args):
 
     librispeech = LibriSpeech(manifest_dir=args.manifest_dir)
 
-    train_cuts = librispeech.train_clean_100_cuts()
     if params.full_libri:
-        train_cuts += librispeech.train_clean_360_cuts()
-        train_cuts += librispeech.train_other_500_cuts()
+        train_cuts = librispeech.train_all_shuf_cuts()
+    else:
+        train_cuts = librispeech.train_clean_100_cuts()
 
     train_cuts = filter_short_and_long_utterances(train_cuts, sp)
 

egs/librispeech/ASR/pruned_transducer_stateless4/train.py

@@ -978,10 +978,10 @@ def run(rank, world_size, args):
 
     librispeech = LibriSpeechAsrDataModule(args)
 
-    train_cuts = librispeech.train_clean_100_cuts()
     if params.full_libri:
-        train_cuts += librispeech.train_clean_360_cuts()
-        train_cuts += librispeech.train_other_500_cuts()
+        train_cuts = librispeech.train_all_shuf_cuts()
+    else:
+        train_cuts = librispeech.train_clean_100_cuts()
 
     def remove_short_and_long_utt(c: Cut):
         # Keep only utterances with duration between 1 second and 20 seconds

egs/librispeech/ASR/pruned_transducer_stateless5/train.py

@@ -1009,10 +1009,10 @@ def run(rank, world_size, args):
 
     librispeech = LibriSpeechAsrDataModule(args)
 
-    train_cuts = librispeech.train_clean_100_cuts()
     if params.full_libri:
-        train_cuts += librispeech.train_clean_360_cuts()
-        train_cuts += librispeech.train_other_500_cuts()
+        train_cuts = librispeech.train_all_shuf_cuts()
+    else:
+        train_cuts = librispeech.train_clean_100_cuts()
 
     def remove_short_and_long_utt(c: Cut):
         # Keep only utterances with duration between 1 second and 20 seconds

egs/librispeech/ASR/pruned_transducer_stateless6/train.py

@@ -970,10 +970,10 @@ def run(rank, world_size, args):
 
     librispeech = LibriSpeechAsrDataModule(args)
 
-    train_cuts = librispeech.train_clean_100_cuts()
     if params.full_libri:
-        train_cuts += librispeech.train_clean_360_cuts()
-        train_cuts += librispeech.train_other_500_cuts()
+        train_cuts = librispeech.train_all_shuf_cuts()
+    else:
+        train_cuts = librispeech.train_clean_100_cuts()
 
     def remove_short_and_long_utt(c: Cut):
         # Keep only utterances with duration between 1 second and 20 seconds

egs/librispeech/ASR/tdnn_lstm_ctc/asr_datamodule.py

@@ -414,6 +414,16 @@ class LibriSpeechAsrDataModule:
             self.args.manifest_dir / "librispeech_cuts_train-other-500.jsonl.gz"
         )
 
+    @lru_cache()
+    def train_all_shuf_cuts(self) -> CutSet:
+        logging.info(
+            "About to get the shuffled train-clean-100, \
+            train-clean-360 and train-other-500 cuts"
+        )
+        return load_manifest_lazy(
+            self.args.manifest_dir / "librispeech_cuts_train-all-shuf.jsonl.gz"
+        )
+
     @lru_cache()
     def dev_clean_cuts(self) -> CutSet:
         logging.info("About to get dev-clean cuts")

egs/librispeech/ASR/tdnn_lstm_ctc/train.py

@@ -173,7 +173,7 @@ def get_params() -> AttributeDict:
         {
             "exp_dir": Path("tdnn_lstm_ctc/exp"),
             "lang_dir": Path("data/lang_phone"),
-            "lr": 1e-3,
+            "lr": 1e-4,
             "feature_dim": 80,
             "weight_decay": 5e-4,
             "subsampling_factor": 3,
@@ -557,10 +557,10 @@ def run(rank, world_size, args):
 
     librispeech = LibriSpeechAsrDataModule(args)
 
-    train_cuts = librispeech.train_clean_100_cuts()
     if params.full_libri:
-        train_cuts += librispeech.train_clean_360_cuts()
-        train_cuts += librispeech.train_other_500_cuts()
+        train_cuts = librispeech.train_all_shuf_cuts()
+    else:
+        train_cuts = librispeech.train_clean_100_cuts()
 
     def remove_short_and_long_utt(c: Cut):
         # Keep only utterances with duration between 1 second and 20 seconds

egs/librispeech/ASR/transducer/train.py

@@ -614,10 +614,10 @@ def run(rank, world_size, args):
 
     librispeech = LibriSpeechAsrDataModule(args)
 
-    train_cuts = librispeech.train_clean_100_cuts()
     if params.full_libri:
-        train_cuts += librispeech.train_clean_360_cuts()
-        train_cuts += librispeech.train_other_500_cuts()
+        train_cuts = librispeech.train_all_shuf_cuts()
+    else:
+        train_cuts = librispeech.train_clean_100_cuts()
 
     def remove_short_and_long_utt(c: Cut):
         # Keep only utterances with duration between 1 second and 20 seconds

egs/librispeech/ASR/transducer_lstm/train.py

@@ -620,10 +620,10 @@ def run(rank, world_size, args):
 
     librispeech = LibriSpeechAsrDataModule(args)
 
-    train_cuts = librispeech.train_clean_100_cuts()
     if params.full_libri:
-        train_cuts += librispeech.train_clean_360_cuts()
-        train_cuts += librispeech.train_other_500_cuts()
+        train_cuts = librispeech.train_all_shuf_cuts()
+    else:
+        train_cuts = librispeech.train_clean_100_cuts()
 
     def remove_short_and_long_utt(c: Cut):
         # Keep only utterances with duration between 1 second and 20 seconds

egs/librispeech/ASR/transducer_stateless/train.py

@@ -641,10 +641,10 @@ def run(rank, world_size, args):
     if params.print_diagnostics:
         diagnostic = diagnostics.attach_diagnostics(model)
 
-    train_cuts = librispeech.train_clean_100_cuts()
     if params.full_libri:
-        train_cuts += librispeech.train_clean_360_cuts()
-        train_cuts += librispeech.train_other_500_cuts()
+        train_cuts = librispeech.train_all_shuf_cuts()
+    else:
+        train_cuts = librispeech.train_clean_100_cuts()
 
     def remove_short_and_long_utt(c: Cut):
         # Keep only utterances with duration between 1 second and 20 seconds

egs/librispeech/ASR/transducer_stateless2/train.py

@@ -629,10 +629,10 @@ def run(rank, world_size, args):
     if params.print_diagnostics:
         diagnostic = diagnostics.attach_diagnostics(model)
 
-    train_cuts = librispeech.train_clean_100_cuts()
     if params.full_libri:
-        train_cuts += librispeech.train_clean_360_cuts()
-        train_cuts += librispeech.train_other_500_cuts()
+        train_cuts = librispeech.train_all_shuf_cuts()
+    else:
+        train_cuts = librispeech.train_clean_100_cuts()
 
     def remove_short_and_long_utt(c: Cut):
         # Keep only utterances with duration between 1 second and 20 seconds

egs/librispeech/ASR/transducer_stateless_multi_datasets/train.py

@@ -752,10 +752,10 @@ def run(rank, world_size, args):
 
     librispeech = LibriSpeech(manifest_dir=args.manifest_dir)
 
-    train_cuts = librispeech.train_clean_100_cuts()
     if params.full_libri:
-        train_cuts += librispeech.train_clean_360_cuts()
-        train_cuts += librispeech.train_other_500_cuts()
+        train_cuts = librispeech.train_all_shuf_cuts()
+    else:
+        train_cuts = librispeech.train_clean_100_cuts()
 
     train_cuts = filter_short_and_long_utterances(train_cuts)