Question wake word possible

[daKmoR]

hey 👋

I'm currently investigating if creating an voice assistant using livekit on an esp32 is "possible".

e.g. I have my livekit demo that runs and I can connect to the room and talk to it my agent... so I assume with this library I could connect an esp32 (with a mic + speaker) to this room as well.

That already sounds pretty awesome.

Question is now... would this always... e.g. 24/7 send an audio stream to the STT system?

my naive ideal solution would be

1. use a wake word like "hey jarvis" to start a conversation => e.g. this would connect to the room
2. have a conversion with live audio transfer as long as there is something to discuss (using EOU https://blog.livekit.io/using-a-transformer-to-improve-end-of-turn-detection/)
3. if "end of conversation" then disconnect from the room and wait for the wake word again

I think this would feel like a totally natural conversion from start to finish 🤞 (instead of having to say alaxa 10 times for a question with some follow ups)

long story short - do you think something like this would be possible with embedded-sdk?

PS: feel free to close this issue if this is not the right place for such a discussion

[daKmoR]

there is https://github.com/kahrendt/microWakeWord which runs on ESP32.

it's usually used via https://esphome.io/

stupid question but building ontop/integrated with esphome would probably be a bigger change I guess? 🤔