Issue Transcribing Twilio Media Stream Using @aws-sdk/client-transcribe-streaming

[ajporterfield]

I'm attempting to transcribe audio coming from a Twilio Media Stream using @aws-sdk/client-transcribe-streaming. I am sending audio to AWS via StartStreamTranscriptionCommand and getting responses back, but they're all empty (ex: { TranscriptEvent: { Transcript: { Results: [] } } }).

The audio from Twilio is coming over a websocket connection in 20ms chucks. The audio is base64 encoded mono, 8-bit 8kHz mulaw. Here is the transform stream I'm writing to.

new stream.Transform({
  transform(chunk, encoding, done) {
    const wav = new WaveFile();
    wav.fromScratch(1, 8000, '8m', Buffer.from(chunk, 'base64'));
    wav.fromMuLaw();
    this.push(Buffer.from(wav.data.samples));
    done();
  }
});

I'm using the wavefile lib to help with the mulaw > pcm conversion. The other pieces involved are pretty much copy and pasted from the code snippets at https://www.npmjs.com/package/@aws-sdk/client-transcribe-streaming.

I'm not sure if the audio encoding is the issue, if it's something related to the transform stream, maybe 20ms chunks are too small. I did see this mentioned on the best practices webpage "PCM (only signed 16-bit little-endian audio formats, which does not include WAV)". I thought a WAV file was just a wrapper for PCM encoding. According to wavefile, fromMulaw() decodes 8-bit mu-Law as 16-bit linear PCM - which sounds like what I need.

I'd really appreciate any help that you could provide to get this working. Thanks!

[ajporterfield]

Quick update, I was able to get this working this morning with a slight variation on the code snipped above.

new stream.Transform({
  transform(chunk, encoding, done) {
    const wav = new WaveFile();
    wav.fromScratch(1, 8000, '8m', chunk);
    wav.fromMuLaw();
    this.push(Buffer.from(wav.data.samples));
    done();
  }
});

This is how I'm writing to the transform stream.

transformStream.write(Buffer.from(payload, 'base64'))

[AdityaPurohit11]

lifesaver thanks a lot @ajporterfield

[github-actions[bot]]

Hello! Reopening this discussion to make it searchable.

[patelaryan0914]

Thanks For the help. Saved a lot of time.