I am receiving all the packets from twilio and also able to send it back and get a response the LLM is giving. I am using STT, TTS and LLM to do my processing. The problem is that when I am doing STT, the STT provider I am using requires the audio_data in wav/mp3 format in 8khz/16khz. Since twilio sends data in mulaw format at 8khz, we need to convert first this into a linear format and then convert it into wav format. I am using audioop.ulaw2lin() method to convert it first into linear then make it into wav format. Because of doing this process, the audio I am getting before sending it to STT is becoming sped up and some words become discernible and very difficult to translate. I have tried reducing sample rate but since twilio sends it at 8000 hz, if I decrease or increase it then its even worse. Can someone help me with this? elif data['event'] == 'media': # Decode base64 μ-law audio from Twilio audio_data = base64.b64decode(data['media']['payload']) # Convert μ-law to PCM pcm_audio = audioop.ulaw2lin(audio_data, 2) audio_data = self.convert_to_wav_8khz(pcm_audio) def convert_to_wav_8khz(self, audio_bytes): """Convert raw audio byte stream to 8kHz WAV (in-memory).""" # Load audio from bytes (pydub auto-detects format) audio = AudioSegment.from_raw( BytesIO(audio_bytes), sample_width=2, frame_rate=8000, channels=1 ) # Convert to mono and 8kHz audio = audio.set_channels(1).set_frame_rate(8000) # Save to an in-memory buffer as WAV wav_buffer = BytesIO() audio.export(wav_buffer, format="wav") return wav_buffer.getvalue() Continue reading...
