To transcribe audio from two speakers using LiveKit AI, here’s a friendly guide to help you through the process: 1. Choose an STT Provider with Diarization Support: It’s great to use providers like Speechmatics or Deepgram that support speaker diarization. For instance, you can enable diarization by setting

enable_diarization=True

when you initialize the STT instance. 2. Use the MultiSpeakerAdapter: Wrapping your STT engine with the

MultiSpeakerAdapter

really helps improve speaker detection. You can configure it to format transcripts with speaker IDs and even choose to suppress background speakers if you like. Here’s an example configuration:

stt = MultiSpeakerAdapter(
    stt=speechmatics.STT(enable_diarization=True),
    detect_primary_speaker=True,
    suppress_background_speaker=False,
    primary_format="<{speaker_id}>{text}",
    background_format="<{speaker_id}>{text}"
)

3. Integrate into an Agent Session: Once set up, use this STT instance within an

AgentSession

to process real-time audio and get transcripts that clearly show who’s speaking. 4. Process Transcripts: Your transcriptions will include speaker labels like

<s1>

and

<s2>

, making it easier to follow the conversation between speakers. 5. Optional: For even better clarity, you can map speaker IDs to participant names using tools like

set_speaker_name

. Just a quick reminder to ensure your STT service supports diarization and that your audio input is a single mixed track. If you have any questions or need a hand with the setup, please don’t hesitate to reach out—I’m here to help! Sources: examples/other/translation/multi-user-translator.py | pipeline-stt/diarization.py