This would directly work if training on Colab/Kaggle. Fixed some issues with the deprecated libraries.

environment.yml

@@ -0,0 +1,13 @@
+name: hifi-gan
+channels:
+  - defaults
+dependencies:
+  - pip
+  - pip:
+    - torch==1.4.0
+    - numpy==1.17.4
+    - librosa==0.7.2
+    - scipy==1.4.1
+    - tensorboard==2.0
+    - soundfile==0.10.3.post1
+    - matplotlib==3.1.3

meldataset.py

@@ -10,42 +10,33 @@
 
 MAX_WAV_VALUE = 32768.0
 
-
 def load_wav(full_path):
     sampling_rate, data = read(full_path)
     return data, sampling_rate
 
-
 def dynamic_range_compression(x, C=1, clip_val=1e-5):
     return np.log(np.clip(x, a_min=clip_val, a_max=None) * C)
 
-
 def dynamic_range_decompression(x, C=1):
     return np.exp(x) / C
 
-
 def dynamic_range_compression_torch(x, C=1, clip_val=1e-5):
     return torch.log(torch.clamp(x, min=clip_val) * C)
 
-
 def dynamic_range_decompression_torch(x, C=1):
     return torch.exp(x) / C
 
-
 def spectral_normalize_torch(magnitudes):
     output = dynamic_range_compression_torch(magnitudes)
     return output
 
-
 def spectral_de_normalize_torch(magnitudes):
     output = dynamic_range_decompression_torch(magnitudes)
     return output
 
-
 mel_basis = {}
 hann_window = {}
 
-
 def mel_spectrogram(y, n_fft, num_mels, sampling_rate, hop_size, win_size, fmin, fmax, center=False):
     if torch.min(y) < -1.:
         print('min value is ', torch.min(y))
@@ -54,24 +45,23 @@ def mel_spectrogram(y, n_fft, num_mels, sampling_rate, hop_size, win_size, fmin,
 
     global mel_basis, hann_window
     if fmax not in mel_basis:
-        mel = librosa_mel_fn(sampling_rate, n_fft, num_mels, fmin, fmax)
+        mel = librosa_mel_fn(sr=sampling_rate, n_fft=n_fft, n_mels=num_mels, fmin=fmin, fmax=fmax)
         mel_basis[str(fmax)+'_'+str(y.device)] = torch.from_numpy(mel).float().to(y.device)
         hann_window[str(y.device)] = torch.hann_window(win_size).to(y.device)
 
     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[str(y.device)],
-                      center=center, pad_mode='reflect', normalized=False, onesided=True)
+                      center=center, pad_mode='reflect', normalized=False, onesided=True, return_complex=True)
 
-    spec = torch.sqrt(spec.pow(2).sum(-1)+(1e-9))
+    spec = torch.abs(spec)
 
     spec = torch.matmul(mel_basis[str(fmax)+'_'+str(y.device)], spec)
     spec = spectral_normalize_torch(spec)
 
     return spec
 
-
 def get_dataset_filelist(a):
     with open(a.input_training_file, 'r', encoding='utf-8') as fi:
         training_files = [os.path.join(a.input_wavs_dir, x.split('|')[0] + '.wav')
@@ -82,7 +72,6 @@ def get_dataset_filelist(a):
                             for x in fi.read().split('\n') if len(x) > 0]
     return training_files, validation_files
 
-
 class MelDataset(torch.utils.data.Dataset):
     def __init__(self, training_files, segment_size, n_fft, num_mels,
                  hop_size, win_size, sampling_rate,  fmin, fmax, split=True, shuffle=True, n_cache_reuse=1,

requirements.txt

@@ -1,7 +1,7 @@
-torch==1.4.0
-numpy==1.17.4
-librosa==0.7.2
-scipy==1.4.1
-tensorboard==2.0
-soundfile==0.10.3.post1
-matplotlib==3.1.3
+torch
+numpy
+librosa
+scipy
+tensorboardX
+soundfile
+matplotlib