Update audio_pipe.py

audio_pipe.py

@@ -106,7 +106,7 @@ class SpeechToSpeechPipeline():
                     [self.tts_model], tts_cfg
                 )
 
-    def __call__(self, inputs: np.array) -> Tuple[np.array, int, List[str]]:
+    def __call__(self, inputs: str) -> Tuple[np.array, int, List[str]]:
         """
         Args:
             inputs (:obj:`np.array`):
@@ -121,7 +121,9 @@ class SpeechToSpeechPipeline():
                     This can be the name of the instruments for audio source separation
                     or some annotation for speech enhancement. The length must be `C'`.
         """
-        _inputs = torch.from_numpy(inputs).unsqueeze(0)
+        # _inputs = torch.from_numpy(inputs).unsqueeze(0)
+        print(f"input: {inputs}")
+        _inputs = torchaudio.load(inputs)
         sample, text = None, None
         if self.cfg.task._name in ["speech_to_text", "speech_to_text_sharded"]:
             sample = S2THubInterface.get_model_input(self.task, _inputs)