How the platform works
A single sermon room. One speaker stream. One translation process. Many listeners.
The speaker signs in and runs a sound check
The speaker selects a microphone, plays a 30-second test, and confirms input volume, noise level, and internet quality before starting.
A LiveKit sermon room is created
The server mints a short-lived access token. The speaker publishes one original-audio track into the room. Listeners subscribe — they do not each start a separate translation.
The translation worker joins the same room
A dedicated translation worker consumes the speaker's audio, auto-detects the language, transcribes, translates to English, generates English speech, and publishes back one translated-audio track and one caption data channel.
Listeners choose English or original
Each listener plays one track at a time — the English translation by default, or the original sermon. Captions render live. Hundreds of listeners share the same outputs.
Recording and retention
The original audio, English audio, and English transcript are recorded. Retention defaults to 30 days. A scheduled backend job removes expired recordings; administrators can extend retention on individual sermon.