Skip to content

Your first event

Who this is for: admins (event producers) — running a live event.

This page is a producer's checklist for a live event with captions. Start with one producer and a short rehearsal before adding microphones or relying on captions for an audience. Captions are generated automatically and can be improved by a human editor, but the result still needs planning. Unfamiliar words are explained in the glossary.

In this stack, a producer operates the capture page while an editor corrects text; an overlay shows the finished captions to the audience. Rooms and links are set up by you or your admin.

Before the event

Test the microphone and audio

  • Connect the audio you will actually use, and pick an input the browser can see: a microphone, a USB audio interface, or a virtual audio device.
  • Speak quietly and normally. Introduce names and any specialist terms, pause, and test background noise.
  • Adjust microphone placement and gain first. If the meter moves but captions do not appear, try Quiet speech sensitivity.
  • Select the spoken language explicitly. Automatic detection is less reliable on short phrases.

Rehearse a sentence

  • Start captions and read a short passage. Pause, then check the text.
  • Run the complete workload for the planned duration. A short successful test does not prove full-event reliability.
  • If the model must be downloaded, do it before arriving at a venue with limited connectivity.
  • Confirm the editor link with the person who will review text. The editor watches the private automatic-caption source room; its published audience room is a different room.
  • In the capture page's sharing section, enter that source room, choose original-language text or English translation, and enable sharing.
  • Confirm corrected captions reach the overlay output you will use, such as OBS, a projector, phones or an embedded frame.
  • Keep source room names private, and use a different source room for each producer whose captions should stay separate.

Check the display and screen layout

  • Open the overlay on the actual audience display, or add it as an OBS Browser Source.
  • Check text size, contrast, line length and display delay on that real screen.
  • If you are screen-sharing, make sure captions do not cover slides or other important content, and that the size still reads well on the stream.
  • Include caption users and accessibility specialists in the rehearsal when possible.

During the event

  • Keep the capture page active and the device awake. Browser suspension, closing the laptop or losing the audio device can stop capture.
  • Monitor the microphone meter, the buffer size and the relay status.
  • Have the editor watch names and exact wording, and correct mistakes.
  • If delay keeps growing, reduce the load: stop unnecessary producers, use transcription-only where appropriate, or move to a configuration you rehearsed. The app stops capture when buffered audio passes thirty seconds and finishes what it already captured.

If something breaks

  • After a connection failure, automatic retries keep the same audio and request identity. Restore the connection, then choose Retry pending audio. Choose Discard only when you intend to lose that audio.
  • For microphone, language or delay problems mid-event, see Troubleshooting.

After the event

  • Choose Stop and wait for processing to finish.
  • Download the transcript before closing the page. Nothing is archived automatically. See Save and share transcripts.
  • Stop the service when everyone has finished.

To stop a terminal service, press Ctrl+C in the window running it. If the stack was started with Docker Compose, stop it with this command:

docker compose down

This keeps downloaded models for the next event; do not add -v unless you mean to delete them.