azalea/VITS-fast-fine-tuning
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix-torch-stft-error-on-gpus-sm-53

Browse Source 
This pull request addresses an issue that arises when executing the finetune_speaker_v2.py script on GPUs with compute capability less than SM_53. The error occurs at line 104 of mel_processing.py, where the torch.stft() function is called with a half data type. To fix this, I updated the data type to float.
This commit is contained in:
Exulan
2023-03-03 16:17:28 +11:00
parent 8137fb9e06
commit b92361b99f
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
mel_processing.py
+2 -2
View File
@@ -101,8 +101,8 @@ def mel_spectrogram_torch(y, n_fft, num_mels, sampling_rate, hop_size, win_size,
y = torch.nn.functional.pad(y.unsqueeze(1), (int((n_fft-hop_size)/2), int((n_fft-hop_size)/2)), mode='reflect')
y = y.squeeze(1)
spec = torch.stft(y, n_fft, hop_length=hop_size, win_length=win_size, window=hann_window[wnsize_dtype_device],
center=center, pad_mode='reflect', normalized=False, onesided=True)
spec = torch.stft(y.float(), n_fft, hop_length=hop_size, win_length=win_size, window=hann_window[wnsize_dtype_device],
center=center, pad_mode='reflect', normalized=False, onesided=True)
spec = torch.sqrt(spec.pow(2).sum(-1) + 1e-6)