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, …