x

tts.py

@@ -210,7 +210,7 @@ class StyleTTS2(nn.Module):
                 # print(x.shape, 'TTS TTS TTS TTS')
                 if x.shape[2] < 100:
                     x = torch.zeros(1, 1, 1000, device=self.device())  # silence if this sentence was empty
-                    
+
             # NORMALIS / Crop Scratch at end (The endingscratch sound is not solved even with nltk.sentence split & punctuation)
             x = x[..., 40:-4000]
             # x /= x.abs().max() + 1e-7   # preserve as torch
@@ -219,10 +219,10 @@ class StyleTTS2(nn.Module):
                 # nohing to vocode
                 x = torch.zeros(1, 1, 1000, device=self.device())
             total.append(x)
-        
+
         # --
-        # total = 1.94 * torch.cat(total, 2)  # 1.94 * Perhaps exceeding -1,1 affects MIMI encode
-        # total /= 1.02 * total.abs().max() + 1e-7
+        total = torch.cat(total, 2)
+        total /= 1.02 * total.abs().max() + 1e-7
         # --
         return total