deepgeek/CosyVoice
2
0
Fork 0
mirror of https://github.com/FunAudioLLM/CosyVoice.git synced 2026-02-04 09:29:25 +08:00
Code Issues Packages Projects Releases Wiki Activity

remove flow_cache

Browse Source
This commit is contained in:
lyuxiang.lx
2025-05-23 12:50:47 +08:00
parent 88f467a8ac
commit 68100c267a
14 changed files with 365 additions and 955 deletions
Download Patch File Download Diff File Expand all files Collapse all files

13
cosyvoice/dataset/processor.py
View File

@@ -159,7 +159,7 @@ def truncate(data, truncate_length=24576, mode='train'):
def compute_fbank(data,
feat_extractor,
token_mel_ratio=2,
token_mel_ratio=0,
mode='train'):
""" Extract fbank
@@ -176,12 +176,11 @@ def compute_fbank(data,
assert 'text_token' in sample
waveform = sample['speech']
feat = feat_extractor(waveform).squeeze(dim=0).transpose(0, 1)
# trim to align speech_token and speech_feat
token_len = 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]
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