Captions, translations and voices are only as good as the sound that reaches InterScribe. This page helps you pick the right way to send it and check it before your session starts.
How sound gets from your room to your audience
- STEP 1A computer in the room
A computer at the sound desk runs the InterScribe agent. It hears the room through a microphone, or through a mixer connected to it.
- STEP 2A room and its channelEach room owns one AV channel: the path its sound travels down. The agent sends to that channel.
- STEP 3A session that listens
A session set in that room takes its sound from the room's channel. The next session in the same room needs nothing set up again.
- STEP 4A check before it mattersTest the audio on the channel before the session, so you find a sound desk that is not routed to InterScribe.
Choose how to send the sound
Go to Rooms & audio, open the Agents tab and select Set up an agent. You choose one of three kinds:
- Desktop app (recommended): a Mac or PC at the desk. It starts with the computer and shows its status in the menu bar. See Set up the desktop agent.
- Command line: a machine that stays in the venue and runs on its own, such as a Raspberry Pi (64-bit Raspberry Pi OS, tested on a Raspberry Pi 5) or a Mac with Apple Silicon. See Run the command-line agent.
- Just the browser: nothing to install. Good for a one-off session or a computer you do not own. See Send audio from a browser.
Setting up an agent needs an admin or organizer.
From OBS, vMix or a hardware encoder
If you already produce a stream, InterScribe can take its sound. On the room's page, select Add an ingress and choose what sends it:
- RTMP, from OBS, vMix or most encoders: you get a server address and a stream key to paste into the encoder.
- WHIP, from newer encoders and browser tools: you get an address and a key too.
- SRT: we pull the stream from an address you give us.
Why we don't recommend it for in-room events
A stream is encoded, sent over the internet and decoded again before InterScribe hears it. Each step holds a little sound back, so captions and AI voices arrive several seconds later than with an agent at the sound desk.
- In the room, listeners hear the speaker live, and the captions and voices lag noticeably behind them.
- Online only, it is fine: viewers already watch a delayed stream, so the extra seconds go unnoticed.
For in-room and interpreted events, send the sound with the desktop, browser or command-line agent instead.
From a mixer or a USB audio interface
If your venue has a sound desk, send its output into the computer that runs the agent, for example through a USB audio interface. Then choose that input as the agent's microphone.
- In the desktop agent, choose it under Audio settings → Microphone.
- If the input carries two channels, Use mono audio combines the microphone's channels into one.
- Unmute the channel that feeds the agent on the sound desk. A muted desk channel sends silence, and InterScribe hears nothing to caption.
Test the audio before the session
On the Agents tab, each agent has a card. Select Test audio on its card, speak at the desk and watch the meter move.
Test audio asks the agents feeding a room's channel to send for a few minutes, whether or not a session is running. You see We can hear it when sound arrives, or No sound heard when nothing does. A channel with nothing sending to it cannot be tested: set up an agent for it first.
In the Operator Console you can also run a sound check from the Source panel. See Standby, sound check and auto-start.
To tell the different delays apart while you are live, see Tell the delays apart and watch them live.
Music and singing
When Detect songs is on for a session, InterScribe tags the captions sung during a song and tells viewers once the room starts singing. While you are live, Music playing in the Operator Console sends listeners a card saying that music is playing.
If the speech recognition service fails
If the speech recognition service in use fails during a session, InterScribe can switch to another one, so captions carry on.