Modified train.py of tedlium3 models (#597)

Author: shcxlee

egs/tedlium3/ASR/pruned_transducer_stateless/train.py

@@ -658,18 +658,8 @@ def remove_short_and_long_utt(c: Cut):
         # Keep only utterances with duration between 1 second and 17 seconds
         return 1.0 <= c.duration <= 17.0
 
-    num_in_total = len(train_cuts)
-
     train_cuts = train_cuts.filter(remove_short_and_long_utt)
 
-    num_left = len(train_cuts)
-    num_removed = num_in_total - num_left
-    removed_percent = num_removed / num_in_total * 100
-
-    logging.info(f"Before removing short and long utterances: {num_in_total}")
-    logging.info(f"After removing short and long utterances: {num_left}")
-    logging.info(f"Removed {num_removed} utterances ({removed_percent:.5f}%)")
-
     train_dl = tedlium.train_dataloaders(train_cuts)
     valid_cuts = tedlium.dev_cuts()
     valid_dl = tedlium.valid_dataloaders(valid_cuts)

egs/tedlium3/ASR/transducer_stateless/train.py

@@ -627,18 +627,8 @@ def remove_short_and_long_utt(c: Cut):
         # Keep only utterances with duration between 1 second and 17 seconds
         return 1.0 <= c.duration <= 17.0
 
-    num_in_total = len(train_cuts)
-
     train_cuts = train_cuts.filter(remove_short_and_long_utt)
 
-    num_left = len(train_cuts)
-    num_removed = num_in_total - num_left
-    removed_percent = num_removed / num_in_total * 100
-
-    logging.info(f"Before removing short and long utterances: {num_in_total}")
-    logging.info(f"After removing short and long utterances: {num_left}")
-    logging.info(f"Removed {num_removed} utterances ({removed_percent:.5f}%)")
-
     train_dl = tedlium.train_dataloaders(train_cuts)
     valid_cuts = tedlium.dev_cuts()
     valid_dl = tedlium.valid_dataloaders(valid_cuts)