Sophia · Real-Time Voice AI

Naturally.

Voice-first conversations that understand context, emotion, and intent — streamed end-to-end over PCM16 with sub-200 ms turn-taking.

Voice detection 40+ languages End-to-end PCM16 Sub-200 ms latency

Conversation

Aria speaking Idle

You · listening Off

FR Mode

When ON, your text or speech transcript is read back with TTS only (no assistant model).

Turn-taking

End-of-utterance delay (before FAQ / model)

800 ms

Partials stream live; the server waits this long after the last finalized STT segment before running FAQ / SLM.

🟢

Conversation state Listening

Turn: listening

Session & audio path

Mic:Off

ID:—

STT:idle

Latency:—

TTS in:0chunks

You → server: mic → audio-worklet.js → PCM16 @16 kHz → WebSocket binary frames.
Bot → you: server TTS → WebSocket binary PCM16 @ TTS rate → AudioContext playback (unlocked on Start call).

Live socket & pipeline log

AUTO-SCROLL ON

Live transcript

—

Talk to AI.

Naturally.

Sign in

Conversation