Show HN: Qiaohu – offline multimodal voice assistant on Snapdragon 8 Gen 2 (github.com)

Built a fully offline Chinese voice assistant that runs entirely on-device (no server, no cloud).

Pipeline: VAD (Silero, 16kHz) → LLM (Gemma 4 2B via LiteRT-LM on Snapdragon GPU) → TTS (sherpa-onnx + matcha-icefall-zh-baker, 22050Hz) → speaker. Barge-in interrupts TTS mid-sentence.

Demo video in the README. Code and full setup instructions on GitHub.