Update app.py

app.py

@@ -76,9 +76,8 @@ async def chat_with_llm(file: UploadFile = File(...)):
     # Load WAV file
     fs, recorded_waveform = wav.read("input_chat.wav")
     os.remove("input_chat.wav")
-    
-    # Convert to bytes and decode
-    waveform_bytes = recorded_waveform.astype(np.uint8).tobytes()
+    recorded_waveform = recorded_waveform.astype(np.float32) / 32767.0
+    waveform_bytes = recorded_waveform.tobytes()
     user_message = ggwave.decode(instance, waveform_bytes)
     
     # Send to LLM
@@ -89,9 +88,23 @@ async def chat_with_llm(file: UploadFile = File(...)):
     llm_response = chat_completion.choices[0].message.content
     
     # Convert response to audio
-    response_waveform = ggwave.encode(instance, llm_response)
+    """Convert text to a WAV audio file using ggwave and return as response."""
+    encoded_waveform = ggwave.encode(input_text.text, protocolId=1, volume=100)
+    
+    # Convert byte data into float32 array
+    waveform_float32 = np.frombuffer(encoded_waveform, dtype=np.float32)
+    
+    # Normalize float32 data to the range of int16
+    waveform_int16 = np.int16(waveform_float32 * 32767)
+    
+    # Save to buffer instead of a file
     buffer = io.BytesIO()
-    wav.write(buffer, 44100, np.frombuffer(response_waveform, dtype=np.uint8))
+    with wave.open(buffer, "wb") as wf:
+        wf.setnchannels(1)                  # Mono audio
+        wf.setsampwidth(2)                  # 2 bytes per sample (16-bit PCM)
+        wf.setframerate(48000)              # Sample rate
+        wf.writeframes(waveform_int16.tobytes())  # Write waveform as bytes
+    
     buffer.seek(0)
     
     return Response(content=buffer.getvalue(), media_type="audio/wav", headers={