I built a 100% local, CPU-only voice loop for any LLM — no GPU, no cloud, nothing leaves your machine (Silero VAD + Parakeet STT + Supertonic TTS 3)
Every voice interface I found either needed a GPU, a cloud API, or was locked to one OS. So I built one that needs none of that — and benchmarked it so the numbers are real. The stack — all ONNX, all CPU: Silero VAD — neural voice activity detection, …