Hey man, should we use 32768.0 instead of 255.0 ?
@app.post("/predict")
async def predict(audio_data: bytes = Depends(parse_body)):
# Convert the audio bytes to a NumPy array
# audio_data_array: np.ndarray = np.frombuffer(audio_data, np.int16).astype(np.float32) / 255.0
audio_data_array: np.ndarray = np.frombuffer(audio_data, np.int16).astype(np.float32) / 32768.0