Speaker Diarization / Identification for Microphone-only Meetings

[SurfMJ]

Currently, when using the "Stream Audio" tab to transcribe in-person meetings with just a microphone (without a web-based meeting platform), all speakers are attributed to a single participant name. This makes it difficult to distinguish who said what in multi-person in-person meetings.

LMA could enable Transcribe's native speaker diarization feature for Stream Audio mode to at least distinguish between different speakers (spk_0, spk_1, etc.), even without knowing their actual names.

[rstrahan]

Agreed, and in fact this is on our roadmap for an upcoming release.
We also hope to leverage this capability for ambient listening / conference room scenarios as well.
And, if participants identify themselves early in a call we can use LLM inferences to assign names to speaker labels.

I am curious about your use case.. Are you using Stream Audio for meeting platform that LMA doesn't (yet) support in Virtual Participant or Chrome Extension? Can you share which meeting platform? Would a better solution (for you) be to extend Virtual Participant and/or Chrome Extension to support your meeting vendor platform?
Or is your use case for something different?

cc @jer2f