endpointing

Enterprise-grade browser extension bringing multilingual voice interaction to AI chatbots (Pi, Claude, ChatGPT). Features real-time speech detection with Silero VAD, accurate transcription via OpenAI Whisper, and ElevenLabs TTS. Built with TypeScript, XState, and modern web standards. Progressive enhancement across Chrome, Firefox, and Safari

Enterprise-grade browser extension bringing multilingual voice interaction to AI chatbots (Pi, Claude, ChatGPT). Features real-time speech detection with Silero VAD, accurate transcription via OpenAI Whisper, and ElevenLabs TTS. Built with TypeScript, XState, and modern web standards. Progressive enhancement across Chrome, Firefox, and Safari

Ends a voice-agent turn on silence measured in audio-stream time, not wall clock

End-of-utterance (turn) detection that rides along with SenseVoice ASR — one forward pass gives you both the transcript and P(end of turn). No extra model, runs on a Raspberry Pi.

Predict the silence as a distribution, then decide when to speak. A model-agnostic turn-taking timing layer for spoken dialogue: calibrated silence-survival forecasting + optimal stopping under an asymmetric cost.

🔄 Adaptive semantic endpointing, four-state conversational machine, and sub-millisecond Barge-In controller for Java voice agents.