I wanted the experience to be 1) local 2) realtime and 3) with voice cloning. Under the hood it’s VAD -> STT -> LLM -> TTS, respectively: - Silero - Parakeet CTC 0.6B - LFM2.5-350 ⁃ Pocket TTS
Ofc, they're not technically your thoughts. I finetuned the LLM on Simone Weil’s and Albert Camus’s personal notebooks. It is a small model, dumb in a weird fun way. You can also finetune it on your own notes or messages, and the experience with both your voice and some of your thought-pieces is a bit uncanny. Here’s the repo if you’d like to try: https://github.com/vidalmaxime/voices-in-your-head