Cloyir/index-tts2-ForDgxSpark
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix: 修复样本音频太长报错的问题，对音频进行裁切。

Browse Source
This commit is contained in:
LGZwr 2025-08-19 16:53:29 +08:00 committed by nanaoto
parent d777b8a029
commit 2cfc76ad9c
1 changed files with 9 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
indextts/infer.py
View File

@ -305,6 +305,15 @@ class IndexTTS:
if audio.shape[0] > 1: if audio.shape[0] > 1:
audio = audio[0].unsqueeze(0) audio = audio[0].unsqueeze(0)
audio = torchaudio.transforms.Resample(sr, 24000)(audio) audio = torchaudio.transforms.Resample(sr, 24000)(audio)
max_audio_length_seconds = 50
max_audio_samples = int(max_audio_length_seconds * 24000)
if audio.shape[1] > max_audio_samples:
if verbose:
print(f"Audio too long ({audio.shape[1]} samples), truncating to {max_audio_samples} samples")
audio = audio[:, :max_audio_samples]
cond_mel = MelSpectrogramFeatures()(audio).to(self.device) cond_mel = MelSpectrogramFeatures()(audio).to(self.device)
cond_mel_frame = cond_mel.shape[-1] cond_mel_frame = cond_mel.shape[-1]
if verbose: if verbose: