Adding speaker recognition to Assistant Pipeline can improve usability of voice assistant.
Maybe it can be run as add-on running with Whisper.
Make speaker recognition part of Wyoming protocol.
I’d like to bring this up again. Diarization or speaker idenfitication are a really cool thing in Google Home devices, where I can work with my own calendar and todo list while my wife works with her services.
Similarly we could allow creating speech embeddings of multiple people within Home Assistant. Each speaker could be mapped to a specific Assist pipeline, which then has only the devices and services enabled, which are mapped to this person.
Similarly, specific prompts and specific MCP servers could be enabled.
Diart could be one of many helpful libraries to achieve this.
5 Likes
I second this. I’ll take just the “speaker identity” as part of the intent trigger.
2 Likes
Any progress on speaker recognition in whisper or home assistant?
1 Like
I would like this as well
yes please! I need that as well.
Definitely. Another use case is user presence detection. If a voice-print is linked to a user in a similar manner as their Wi-Fi device’s MAC address, HA can know if that user is present if it hears their voice, even if they don’t have their smartphone in hand or if it has GPS, Wi-Fi, Bluetooth off.
Important: Feature requests have moved to home-assistant · Discussions · GitHub so if OP (or any others on this thread if OP busy) can enter it there with the use cases & link back to this thread, it’ll be seen more. I’ll be sure to vote on it there.
1 Like
Hi, I have created a feature request in the GitHub Discussions.