Merge pull request #369 from kan-bayashi/fix_extream_slow

parallel_wavegan/bin/train.py

@@ -597,6 +597,7 @@ def __call__(self, batch):
         c_batch = [c[start:end] for c, start, end in zip(cs, c_starts, c_ends)]
 
         # convert each batch to tensor, asuume that each item in batch has the same length
+        y_batch, c_batch = np.array(y_batch), np.array(c_batch)
         y_batch = torch.tensor(y_batch, dtype=torch.float).unsqueeze(1)  # (B, 1, T)
         c_batch = torch.tensor(c_batch, dtype=torch.float).transpose(2, 1)  # (B, C, T')