diff --git a/egs/libriheavy/ASR/zipformer_prompt_asr/dataset.py b/egs/libriheavy/ASR/zipformer_prompt_asr/dataset.py index 47bbfea41..f808ac14d 100644 --- a/egs/libriheavy/ASR/zipformer_prompt_asr/dataset.py +++ b/egs/libriheavy/ASR/zipformer_prompt_asr/dataset.py @@ -238,7 +238,7 @@ def triplet_text_sampling( lower_all_char, ] - sampling_weight = [0.5, 0.2, 0.15, 0.15] # Mixed-punc should have the largest sampling prob + sampling_weight = [0.7, 0.3, 0.0, 0.0] # Mixed-punc should have the largest sampling prob total_transforms = len(transforms) # do not use the recognized trans diff --git a/egs/libriheavy/ASR/zipformer_prompt_asr/dataset2.py b/egs/libriheavy/ASR/zipformer_prompt_asr/dataset2.py index 0a37a7d2c..8d0fb4427 100644 --- a/egs/libriheavy/ASR/zipformer_prompt_asr/dataset2.py +++ b/egs/libriheavy/ASR/zipformer_prompt_asr/dataset2.py @@ -240,7 +240,8 @@ def triplet_text_sampling( lower_all_char, ] - sampling_weight = [0.5, 0.2, 0.15, 0.15] # Mixed-punc should have the largest sampling prob + # sampling_weight = [0.5, 0.2, 0.15, 0.15] # Mixed-punc should have the largest sampling prob + sampling_weight = [0.7, 0.3, 0.0, 0.0] total_transforms = len(transforms) # do not use the recognized trans