blackteay/GPT-SoVITS
1
0
Fork 0
mirror of https://github.com/RVC-Boss/GPT-SoVITS.git synced 2025-12-16 17:40:09 +08:00
Code Issues Packages Projects Releases Wiki Activity

modified: GPT_SoVITS/TTS_infer_pack/TTS.py

Browse Source
This commit is contained in:
ChasonJiang 2025-07-01 22:27:03 +08:00
parent 0825ae80e1
commit d08214dd22
1 changed files with 10 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
GPT_SoVITS/TTS_infer_pack/TTS.py
View File

@ -1799,9 +1799,16 @@ class TTS:
f2 = audio_fragments[i + 1] f2 = audio_fragments[i + 1]
w1 = f1[-overlap_len:] w1 = f1[-overlap_len:]
w2 = f2[:overlap_len] w2 = f2[:overlap_len]
assert w1.shape == w2.shape w2 = w2[-w2.shape[-1]//2:]
corr = F.conv1d(w1.view(1, 1, -1), w2.view(1, 1, -1), padding=w2.shape[-1] // 2).view(-1)[:-1] # assert w1.shape == w2.shape
idx = corr.argmax() corr = F.conv1d(w1.view(1, 1, -1), w2.view(1, 1, -1)).view(-1)
squared_sum = F.conv1d(w1.view(1, 1, -1)**2, torch.ones_like(w2).view(1, 1, -1)).view(-1)+ 1e-8
idx = (corr/squared_sum.sqrt()).argmax()
print(f"seg_idx: {idx}")
# idx = corr.argmax()
f1_ = f1[: -(overlap_len - idx)] f1_ = f1[: -(overlap_len - idx)]
audio_fragments[i] = f1_ audio_fragments[i] = f1_