Voice demo

PulseVAD → Moonshine tiny → LLM → Pocket TTS disconnected

Conversation

Click Start, allow the microphone, and talk. Use headphones.

Latency — this session

voice-to-voice, last turn
ms
voice-to-voice p50 / p95
ms
turns
0
#v2vSTTLLM 1st toksentence+TTSspec.you said
v2v: VAD declared you stopped (0.2 s of silence) → first bot audio sent. Add ~0.2–0.3 s for the silence itself to get the pause you hear. STT: Moonshine tiny on the whole utterance. LLM 1st tok: from transcript to first token (OpenRouter). sentence+TTS: first LLM token → first Pocket TTS audio (pipecat waits for the first full sentence before synthesising it). spec.: ✓ the reply was generated speculatively at 0.2 s and released at 0.6 s; ✗ the speculation was discarded (you resumed talking). Server-wide numbers: /metrics.