Fix vocoder interface (#1895)

Author: mthrok

torchaudio/pipelines/_tts/impl.py

@@ -83,7 +83,7 @@ def __init__(
     def sample_rate(self):
         return self._sample_rate
 
-    def forward(self, mel_spec, lengths):
+    def forward(self, mel_spec, lengths=None):
         mel_spec = torch.exp(mel_spec)
         mel_spec = 20 * torch.log10(torch.clamp(mel_spec, min=1e-5))
         if self._min_level_db is not None:
@@ -120,7 +120,7 @@ def __init__(self):
     def sample_rate(self):
         return self._sample_rate
 
-    def forward(self, mel_spec, lengths):
+    def forward(self, mel_spec, lengths=None):
         mel_spec = torch.exp(mel_spec)
         mel_spec = mel_spec.clone().detach().requires_grad_(True)
         spec = self._inv_mel(mel_spec)

torchaudio/pipelines/_tts/interface.py

@@ -47,7 +47,7 @@ def sample_rate(self):
         """
 
     @abstractmethod
-    def __call__(self, specgrams: Tensor, lengths: Optional[Tensor]) -> Tuple[Tensor, Optional[Tensor]]:
+    def __call__(self, specgrams: Tensor, lengths: Optional[Tensor] = None) -> Tuple[Tensor, Optional[Tensor]]:
         """Generate waveform from the given input, such as spectrogram
 
         See :func:`torchaudio.pipelines.Tacotron2TTSBundle.get_vocoder` for the usage.
@@ -58,6 +58,7 @@ def __call__(self, specgrams: Tensor, lengths: Optional[Tensor]) -> Tuple[Tensor
                 The expected shape depends on the implementation.
             lengths (Tensor, or None, optional):
                 The valid length of each sample in the batch. Shape: `(batch, )`.
+                (Default: `None`)
 
         Returns:
             (Tensor, Optional[Tensor]):