I discovered a research paper, I'll try to implement what they demonstrated, and then experiment to make it use as less resource as possible
> DuplexCascade: Full-Duplex Speech-to-Speech Dialogue with VAD-Free Cascaded ASR–LLM–TTS Pipeline and Micro-Turn Optimization
> https://arxiv.org/html/2603.09180
Would be fun to have a full duplex with a local LLM :D
Replying to @lexoyo@framapiaf.org
@lexoyo I'm interested! When you have something to test, please ping me. Everything speech is a topic interesting for me!
Replying to @menelion@caneandable.social
@menelion if you feel like testing, here is the code I have
https://github.com/lexoyo/microturn/
It should work on linux with the instructions in the reame
It's quite impressive how responsive it feels :))