From fe9f975ec2cb577bcde8f4aa42d5c454e881352b Mon Sep 17 00:00:00 2001 From: Bailey Hirota Date: Fri, 13 Jun 2025 00:48:37 +0900 Subject: [PATCH] changes to train script - no need for limiting utterance length here --- egs/multi_ja_en/ASR/local/validate_bpe_lexicon.py | 2 +- egs/multi_ja_en/ASR/zipformer/train.py | 14 +++----------- 2 files changed, 4 insertions(+), 12 deletions(-) diff --git a/egs/multi_ja_en/ASR/local/validate_bpe_lexicon.py b/egs/multi_ja_en/ASR/local/validate_bpe_lexicon.py index 4e843acf5..f17e1cc6d 120000 --- a/egs/multi_ja_en/ASR/local/validate_bpe_lexicon.py +++ b/egs/multi_ja_en/ASR/local/validate_bpe_lexicon.py @@ -1 +1 @@ -/root/icefall/egs/librispeech/ASR/local/validate_bpe_lexicon.py \ No newline at end of file +/root/Github/reazon-icefall/egs/librispeech/ASR/local/validate_bpe_lexicon.py \ No newline at end of file diff --git a/egs/multi_ja_en/ASR/zipformer/train.py b/egs/multi_ja_en/ASR/zipformer/train.py index e3e7bfaf2..c4aaa17db 100755 --- a/egs/multi_ja_en/ASR/zipformer/train.py +++ b/egs/multi_ja_en/ASR/zipformer/train.py @@ -1185,15 +1185,12 @@ def run(rank, world_size, args): train_cuts = multi_dataset.train_cuts() def remove_short_and_long_utt(c: Cut): - # Keep only utterances with duration between 1 second and 30 seconds - # - # Caution: There is a reason to select 30.0 here. Please see - # ../local/display_manifest_statistics.py + # Keep only utterances greater than 1 second # # You should use ../local/display_manifest_statistics.py to get # an utterance duration distribution for your dataset to select - # the threshold - if c.duration < 1.0 or c.duration > 30.0: + # the threshold as this is dependent on which datasets you choose + if c.duration < 1.0: logging.warning( f"Exclude cut with ID {c.id} from training. Duration: {c.duration}" ) @@ -1239,14 +1236,10 @@ def run(rank, world_size, args): else: sampler_state_dict = None - # train_dl = reazonspeech_corpus.train_dataloaders( - # train_cuts, sampler_state_dict=sampler_state_dict - # ) train_dl = multidataset_datamodule.train_dataloaders( train_cuts, sampler_state_dict=sampler_state_dict ) - valid_cuts = multi_dataset.dev_cuts() valid_dl = multidataset_datamodule.valid_dataloaders(valid_cuts) @@ -1393,7 +1386,6 @@ def main(): MultiDatasetAsrDataModule.add_arguments(parser) args = parser.parse_args() args.exp_dir = Path(args.exp_dir) - print(args) world_size = args.world_size assert world_size >= 1