deepgeek/CosyVoice
2
0
Fork 0
mirror of https://github.com/FunAudioLLM/CosyVoice.git synced 2026-02-05 18:09:24 +08:00
Code Issues Packages Projects Releases Wiki Activity

[feature] modify pad to trim

Browse Source 
Conflicts:
	cosyvoice/dataset/processor.py
This commit is contained in:
burkliu
2025-04-25 10:31:43 +08:00
committed by lyuxiang.lx
parent 97f0bc61cd
commit fbab274b6a
1 changed files with 8 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
cosyvoice/dataset/processor.py
View File

@@ -159,6 +159,7 @@ def truncate(data, truncate_length=24576, mode='train'):
def compute_fbank(data, def compute_fbank(data,
feat_extractor, feat_extractor,
token_mel_ratio=0,
mode='train'): mode='train'):
""" Extract fbank """ Extract fbank
@@ -174,8 +175,13 @@ def compute_fbank(data,
assert 'utt' in sample assert 'utt' in sample
assert 'text_token' in sample assert 'text_token' in sample
waveform = sample['speech'] waveform = sample['speech']
mat = feat_extractor(waveform).squeeze(dim=0).transpose(0, 1) feat = feat_extractor(waveform).squeeze(dim=0).transpose(0, 1)
sample['speech_feat'] = mat if token_mel_ratio != 0:
# trim to align speech_token and speech_feat
token_len = int(min(feat.shape[0] / token_mel_ratio, sample["speech_token"].shape[0]))
feat = feat[:token_mel_ratio * token_len]
sample["speech_token"] = sample["speech_token"][:token_len]
sample['speech_feat'] = feat
yield sample yield sample