Update handler.py

handler.py

@@ -38,26 +38,15 @@ class EndpointHandler:
             torch.Tensor: Preprocessed audio tensor ready for inference
         """
         try:
+            audio = input_data if isinstance(input_data, bytes) else input_data['inputs']
             # Handle different input types
-            if isinstance(input_data, bytes):
+            if isinstance(audio, bytes):
                 # Load audio from bytes
-                audio_buffer = BytesIO(input_data)
+                audio_buffer = BytesIO(audio)
                 waveform, sample_rate = torchaudio.load(audio_buffer)
-            elif isinstance(input_data, dict):
-                logger.info(f"Input data: {input_data.keys()}")
-                if 'audio' in input_data:
-                    # Handle numpy array input
-                    audio_array = input_data['audio']
-                    if isinstance(audio_array, list):
-                        audio_array = np.array(audio_array)
-                    waveform = torch.from_numpy(audio_array)
-                    sample_rate = input_data.get('sampling_rate', self.target_sample_rate)
-                    # Ensure 2D tensor [channels, time]
-                    if waveform.dim() == 1:
-                        waveform = waveform.unsqueeze(0)
-                else:
-                    raise ValueError("Input dictionary must contain 'audio' key")
             else:
+                logger.error(f"Unsupported input type: {type(audio)}")
+                logger.debug(f"Input data: {input_data.keys()}")
                 raise ValueError("Unsupported input type")
 
             # Convert to float32