RevOS Voice+ Live Lab

Co-located turn engine · Whisper → local LLM → Kokoro
connecting…
First Audio · full turn
— ms
best: — · avg: —
LLM 1st token
—
ms
1st clause
—
ms
RAG
—
ms
Type a message or hold the mic to talk.
hold
Voice
Endpoint
Lab notes. Audio out is 24 kHz mono PCM, streamed as the vocoder produces it. Mic in is downsampled to 16 kHz for STT. This page connects over ws:// — fine from http://localhost. A public https deploy needs wss:// (TLS) on the box. First turn after the engine is idle is a cold start; steady-state is the real number.