Update app.py

app.py

@@ -74,7 +74,7 @@ def decode_audio(encoded_file_path):
         return f"Decoding error: {e}"
 
 @spaces.GPU(duration=180)
-def stream_decode_audio(encoded_file_path) -> Generator[np.ndarray, None, None]:
+def stream_decode_audio(encoded_file_path) -> Generator[tuple, None, None]:
     try:
         # Load encoded data from the .owie file
         with open(encoded_file_path, 'rb') as temp_file:
@@ -85,15 +85,18 @@ def stream_decode_audio(encoded_file_path) -> Generator[np.ndarray, None, None]:
 
         # Decode the audio in chunks
         chunk_size = 16000  # 1 second of audio at 16kHz
+        sample_rate = 16000  # AGC model's output sample rate
         with torch.no_grad():
             for i in range(0, z.shape[2], chunk_size):
                 z_chunk = z[:, :, i:i+chunk_size]
                 audio_chunk = agc.decode(z_chunk)
-                yield audio_chunk.squeeze(0).cpu().numpy()
+                # Convert to numpy, then to list, and yield as tuple with sample rate
+                audio_data = audio_chunk.squeeze(0).cpu().numpy().T.tolist()
+                yield (sample_rate, audio_data)
 
     except Exception as e:
-        yield np.zeros((2, chunk_size))  # Return silence in case of error
         print(f"Streaming decoding error: {e}")
+        yield (sample_rate, [[0] * chunk_size, [0] * chunk_size])  # Return silence in case of error
 
 # Gradio Interface
 with gr.Blocks() as demo: