With the below code, the result ends up just being always the same "Thank you." Any ideas what could be going wrong. For reference, I used the load_audio function in the whisper package. And an article about twilio and Vosk. @sock.route('/stream') def stream(ws): while True: message = ws.receive() packet = json.loads(message) if packet['event'] == 'media': # get audio as ulaw audio = base64.b64decode(packet['media']['payload']) add_audio(audio) buffer = np.array([], dtype=np.float32) def add_audio(audio): global buffer # convert audio to numpy array audio = np.frombuffer(audio, np.int16).flatten().astype(np.float32) / 32768.0 buffer = np.concatenate((buffer, audio)) process_audio() def process_audio(): global buffer audio = whisper.pad_or_trim(buffer) mel = whisper.log_mel_spectrogram(audio).to(model.device) mel = torch.nan_to_num(mel) result = whisper.decode(model, mel, options) Continue reading...
