Releaseperception2026-07-07
Fully offline voice — no internet required for speech
Speech recognition, voice synthesis, and wake-word detection all run locally. Your robot works the same in a basement or a field.
Full Conversations Without an Internet Connection
Previous versions relied on cloud speech services. Now all voice capabilities run on-device — works perfectly with no network at all.
Offline Capabilities
- Speech Recognition (ASR) — Sherpa-ONNX engine, supports Chinese and English, SenseVoice-Small available as an option
- Text-to-Speech (TTS) — Matcha engine, faster and better sounding than the previous VITS
- Wake Word (KWS) — Say a keyword to wake the robot without continuous listening
Three Trigger Modes
- Always listening — Best for demos and exhibitions
- Wake word — For everyday use, say "hello" to start a conversation
- Push to talk — Manual control, most resource-efficient
Real-World Experience
- Models auto-download to /models on first boot, then run fully offline
- Audio pre-buffering ensures the first syllable is not swallowed by VAD
- 500ms silence padding at the end prevents sentence truncation
Choose your trigger mode and ASR model in Settings.