This is source code for my pet project AI Voice Agent. Idea behind this agent is to have local hotword enabled voice assistant that can answer your questions using blazingly fast OpenAI Realtime API during our podcast recordings.
By default it does use russian language as a primary language, but it can be easily changed to any other language supported.
- Blazingly fast responses through websockets
- Ability to interrupt current response with new question
- Ability to shutdown agent by voice command
Everything should be pretty straightforward. You will need to download navec embeddings and put them into root folder.
- VAD with Silero VAD
- STT with Vosk
- Words embeddings with Navec
- STS with OpenAI Realtime API