DynamicBucketingSampler

egs/gigaspeech/ASR/conformer_ctc/asr_datamodule.py

@@ -22,9 +22,9 @@ from pathlib import Path
 
 from lhotse import CutSet, Fbank, FbankConfig, load_manifest
 from lhotse.dataset import (
-    BucketingSampler,
     CutConcatenate,
     CutMix,
+    DynamicBucketingSampler,
     K2SpeechRecognitionDataset,
     PrecomputedFeatures,
     SingleCutSampler,
@@ -82,7 +82,7 @@ class GigaSpeechAsrDataModule:
         group.add_argument(
             "--bucketing-sampler",
             type=str2bool,
-            default=False,
+            default=True,
             help="When enabled, the batches will come from buckets of "
             "similar duration (saves padding frames).",
         )
@@ -90,7 +90,7 @@ class GigaSpeechAsrDataModule:
             "--num-buckets",
             type=int,
             default=30,
-            help="The number of buckets for the BucketingSampler"
+            help="The number of buckets for the DynamicBucketingSampler"
             "(you might want to increase it for larger datasets).",
         )
         group.add_argument(
@@ -142,7 +142,7 @@ class GigaSpeechAsrDataModule:
         group.add_argument(
             "--num-workers",
             type=int,
-            default=20,
+            default=2,
             help="The number of training dataloader workers that "
             "collect the batches.",
         )
@@ -270,13 +270,12 @@ class GigaSpeechAsrDataModule:
             )
 
         if self.args.bucketing_sampler:
-            logging.info("Using BucketingSampler.")
-            train_sampler = BucketingSampler(
+            logging.info("Using DynamicBucketingSampler.")
+            train_sampler = DynamicBucketingSampler(
                 cuts_train,
                 max_duration=self.args.max_duration,
                 shuffle=self.args.shuffle,
                 num_buckets=self.args.num_buckets,
-                bucket_method="equal_duration",
                 drop_last=True,
             )
         else:
@@ -321,7 +320,7 @@ class GigaSpeechAsrDataModule:
                 cut_transforms=transforms,
                 return_cuts=self.args.return_cuts,
             )
-        valid_sampler = BucketingSampler(
+        valid_sampler = DynamicBucketingSampler(
             cuts_valid,
             max_duration=self.args.max_duration,
             shuffle=False,
@@ -345,7 +344,7 @@ class GigaSpeechAsrDataModule:
             else PrecomputedFeatures(),
             return_cuts=self.args.return_cuts,
         )
-        sampler = BucketingSampler(
+        sampler = DynamicBucketingSampler(
             cuts, max_duration=self.args.max_duration, shuffle=False
         )
         logging.debug("About to create test dataloader")

egs/gigaspeech/ASR/local/compute_fbank_gigaspeech_dev_test.py

@@ -72,9 +72,13 @@ def compute_fbank_gigaspeech_dev_test():
             batch_duration=batch_duration,
             storage_type=LilcomHdf5Writer,
         )
+        cut_set = cut_set.trim_to_supervisions(
+            keep_overlapping=False, min_duration=None
+        )
 
         logging.info(f"Saving to {cuts_path}")
         cut_set.to_file(cuts_path)
+        logging.info(f"Saved to {cuts_path}")
 
 
 def main():