blackteay/GPT-SoVITS
1
0
Fork 0
mirror of https://github.com/RVC-Boss/GPT-SoVITS.git synced 2025-09-29 17:10:02 +08:00
Code Issues Packages Projects Releases Wiki Activity

fixed resample audio and tested in full run

Browse Source
This commit is contained in:
zpeng11 2025-08-20 16:47:55 -04:00
parent da5aa78224
commit 1cdd41d877
2 changed files with 8 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

15
GPT_SoVITS/onnx_export.py
View File

@ -51,14 +51,13 @@ def resample_audio(audio: torch.Tensor, orig_sr: int, target_sr: int) -> torch.T
audio = audio.unsqueeze(0) audio = audio.unsqueeze(0)
# audio shape: (batch, channels, samples) # audio shape: (batch, channels, samples)
batch, channels, samples = audio.shape batch, channels, samples = audio.shape
new_samples = int(samples * target_sr / orig_sr) # Reshape to combine batch and channels for interpolation
audio = audio.view(batch * channels, 1, samples) audio = audio.reshape(batch * channels, 1, samples)
resampled = F.interpolate(audio, size=new_samples, mode='linear', align_corners=False) # Use scale_factor instead of a computed size for ONNX export compatibility
resampled = resampled.view(batch, channels, new_samples) resampled = F.interpolate(audio, scale_factor=target_sr / orig_sr, mode='linear', align_corners=False)
if resampled.shape[0] == 1 and resampled.shape[1] == 1: new_samples = resampled.shape[-1]
resampled = resampled.squeeze(0).squeeze(0) resampled = resampled.reshape(batch, channels, new_samples)
elif resampled.shape[0] == 1: resampled = resampled.squeeze(0).squeeze(0)
resampled = resampled.squeeze(0)
return resampled return resampled

2
playground/freerun.py
View File

@ -7,7 +7,7 @@ import torch
from TTS_infer_pack.TextPreprocessor_onnx import TextPreprocessorOnnx from TTS_infer_pack.TextPreprocessor_onnx import TextPreprocessorOnnx
MODEL_PATH = "playground/v2pro_export/v2pro" MODEL_PATH = "onnx/v2pro_export/v2pro"
def audio_postprocess( def audio_postprocess(
audios, audios,