Informal voice assistant
I took talk-llama and added:
- XTTSv2 streaming.
- low latencies.
- Russian and other languages, UTF-8.
- voice commands: Google, stop, regenerate, reset.
Under the hood:
- STT: whisper.cpp medium
- LLM: Mistral-7B-v0.2-Q6_k.gguf
- TTS: XTTSv2 in streaming mode
- Google: langchain google-serper
In this video I used nvidia 3060 12 GB, but I guess 8 GB of VRAM is also enough. Have plans to port everything to android.
Code, exe, manual:
github.com/Mozer/talk-llama-fast
Негізгі бет Talk llama fast - informal voice assistant [en]
Пікірлер: 5