From f6bae95ebdde0fb0deb2e7513ef55e3c984a8c14 Mon Sep 17 00:00:00 2001 From: Bailey Hirota Date: Tue, 1 Jul 2025 21:21:25 +0900 Subject: [PATCH] commenting --- egs/multi_ja_en/ASR/local/utils/asr_datamodule.py | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-) diff --git a/egs/multi_ja_en/ASR/local/utils/asr_datamodule.py b/egs/multi_ja_en/ASR/local/utils/asr_datamodule.py index 1aa74119f..4a5dd520d 100644 --- a/egs/multi_ja_en/ASR/local/utils/asr_datamodule.py +++ b/egs/multi_ja_en/ASR/local/utils/asr_datamodule.py @@ -206,6 +206,10 @@ class MultiDatasetAsrDataModule: else: logging.info("Disable MUSAN") + # Cut concatenation should be the first transform in the list, + # so that if we e.g. mix noise in, it will fill the gaps between + # different utterances. + if self.args.concatenate_cuts: logging.info( f"Using cut concatenation with duration factor " @@ -298,7 +302,7 @@ class MultiDatasetAsrDataModule: batch_size=None, num_workers=self.args.num_workers, persistent_workers=True, - worker_init_fn=worker_init_fn, #changed bottom 2 params + worker_init_fn=worker_init_fn, ) return train_dl