Merge remote-tracking branch 'origin/main'

finetune_speaker_v2.py

@@ -65,7 +65,8 @@ def run(rank, n_gpus, hps):
     writer = SummaryWriter(log_dir=hps.model_dir)
     writer_eval = SummaryWriter(log_dir=os.path.join(hps.model_dir, "eval"))
 
-  dist.init_process_group(backend='nccl', init_method='env://', world_size=n_gpus, rank=rank)
+  # Use gloo backend on Windows for Pytorch
+  dist.init_process_group(backend=  'gloo' if os.name == 'nt' else 'nccl', init_method='env://', world_size=n_gpus, rank=rank)
   torch.manual_seed(hps.train.seed)
   torch.cuda.set_device(rank)
 
@@ -317,4 +318,4 @@ def evaluate(hps, generator, eval_loader, writer_eval):
 
 
 if __name__ == "__main__":
-  main()
+  main()

mel_processing.py

@@ -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)
 

requirements.txt

@@ -1,5 +1,5 @@
 Cython
-librosa
+librosa==0.9.1
 numpy
 scipy
 tensorboard