new commits

demo_cli.py

@@ -311,7 +311,7 @@ if __name__ == '__main__':
         # Synthesizing the waveform is fairly straightforward. Remember that the longer the
         # spectrogram, the more time-efficient the vocoder.
         if not args.griffin_lim:
-            wav = vocoder.infer_waveform(spec, target=4000, overlap=400)
+            wav = vocoder.infer_waveform(spec, target=vocoder.hp.voc_target, overlap=vocoder.hp.voc_overlap, crossfade=vocoder.hp.is_crossfade) 
         else:
             wav = Synthesizer.griffin_lim(spec)
 

vocoder/hparams.py

@@ -40,8 +40,9 @@ voc_seq_len = hop_length * 5        # must be a multiple of hop_length
 
 # Generating / Synthesizing
 voc_gen_batched = True              # very fast (realtime+) single utterance batched generation
-voc_target = 8000                   # target number of samples to be generated in each batch entry
+voc_target = 4000                   # target number of samples to be generated in each batch entry
 voc_overlap = 400                   # number of samples for crossfading between batches
+is_crossfade = True                 # crossfading or not
 
 # Output Noise Reduce
 prop_decrease_low_freq = 0.6        # prop decrease for low dominant frequency

vocoder/inference.py

@@ -50,7 +50,7 @@ def is_loaded():
 
 
 def infer_waveform(mel, normalize=True,  batched=True, target=8000, overlap=800, 
-                   progress_callback=None):
+                   progress_callback=None, crossfade=True):
     """
     Infers the waveform of a mel spectrogram output by the synthesizer (the format must match 
     that of the synthesizer!)
@@ -67,7 +67,7 @@ def infer_waveform(mel, normalize=True,  batched=True, target=8000, overlap=800,
     if normalize:
         mel = mel / hp.mel_max_abs_value
     mel = torch.from_numpy(mel[None, ...])
-    wav = _model.generate(mel, batched, target, overlap, hp.mu_law, progress_callback)
+    wav = _model.generate(mel, batched, target, overlap, hp.mu_law, progress_callback, crossfade=crossfade)
     wav = waveform_denoising(wav)
     return wav
 

vocoder/models/fatchord_version.py

@@ -150,7 +150,7 @@ class WaveRNN(nn.Module):
         x = F.relu(self.fc2(x))
         return self.fc3(x)
 
-    def generate(self, mels, batched, target, overlap, mu_law, progress_callback=None):
+    def generate(self, mels, batched, target, overlap, mu_law, progress_callback=None,crossfade=True):
         mu_law = mu_law if self.mode == 'RAW' else False
         progress_callback = progress_callback or self.gen_display
 
@@ -240,7 +240,7 @@ class WaveRNN(nn.Module):
         output = output.astype(np.float64)
         
         if batched:
-            output = self.xfade_and_unfold(output, target, overlap)
+            output = self.xfade_and_unfold(output, target, overlap, crossfade=crossfade)
         else:
             output = output[0]
 
@@ -340,7 +340,7 @@ class WaveRNN(nn.Module):
 
         return folded
 
-    def xfade_and_unfold(self, y, target, overlap):
+    def xfade_and_unfold(self, y, target, overlap, crossfade=True):
 
         ''' Applies a crossfade and unfolds into a 1d array.
 
@@ -382,9 +382,12 @@ class WaveRNN(nn.Module):
         silence = np.zeros((silence_len), dtype=np.float64)
 
         # Equal power crossfade
-        t = np.linspace(-1, 1, fade_len, dtype=np.float64)
-        fade_in = np.sqrt(0.5 * (1 + t))
-        fade_out = np.sqrt(0.5 * (1 - t))
+        if crossfade:
+            t = np.linspace(-1, 1, fade_len, dtype=np.float64)
+            fade_in = np.sqrt(0.5 * (1 + t))
+            fade_out = np.sqrt(0.5 * (1 - t))
+        else:
+            fade_in = fade_out = np.ones((fade_len), dtype=np.float64)
 
         # Concat the silence to the fades
         fade_in = np.concatenate([silence, fade_in])