Church audio setup for live translation: a clean feed from mixer to captions
How to get clean speech into live translation: an aux or matrix send to a USB audio interface or an encoder stream, mic technique, levels, what to leave out, a line check and fixes for common faults.

What is the best church audio setup for live translation?
The best church audio setup for live translation is a dedicated speech-only feed: an aux or matrix send from the mixer, carrying only the speech microphones, into a USB audio interface on the operator's computer. Churches that already stream can send audio from OBS, vMix or ProPresenter over RTMP instead. A browser microphone works in small rooms but picks up echo and music.
Speech recognition is the first link in the chain. If it hears a muddy, echoing or clipped signal, every caption and every language downstream inherits the problem, and no translation setting can repair words that never arrived clearly. An hour spent on the feed usually does more for caption quality than anything else you can change.
Which audio path should you use?
| Path | What you need | Strength | Watch out for | Plan |
|---|---|---|---|---|
| Aux or matrix send to a USB interface | A free mixer output, a two-channel USB interface, a balanced cable | Speech only, independent of the room and stream mixes | Gain staging and ground-loop hum | Any plan |
| Digital mixer's built-in USB audio | A USB cable from the desk to the computer | No extra box | Browsers often read only the first channels of a multichannel device, so put the send on USB 1 and 2 | Any plan |
| Encoder input over RTMP or SRT | Your OBS, vMix or ProPresenter computer | No new cabling if you already stream | Carries the stream mix, music and ambience included | Advanced and Premium |
| Browser microphone | A laptop or phone near the speaker | Fast to start; fine for a test or a small room | Room echo, crowd noise and music bleed | Any plan |
For the encoder route, the SermonLingo console shows step-by-step encoder settings; in ProPresenter, use Capture → Custom RTMP destination. If your stream mix leans on room mics for atmosphere, a separate mixer send will still give cleaner captions. Our guide to church livestream translation covers the streaming side.
How do you build a translation send on the mixer?
On an analog mixer
- Find a free aux send. Monitor auxes are usually pre-fader and effects auxes post-fader; many boards have a pre/post switch per aux.
- Turn that aux up only on speech channels: preacher's headset or lapel, pulpit, readers' mic, host mic and the handheld for testimonies.
- Set the aux master near unity (the 0 or U mark) and adjust the channel sends until normal speech reads well on the interface meter.
- Run a balanced cable (TRS or XLR, depending on the output) from the aux output to the interface's line input.
On a digital mixer
- Assign a spare mix bus or matrix and name it TRANSLATION on the scribble strip.
- Send speech channels to it, and nothing else. If your speech mics already feed a subgroup, a matrix fed from that subgroup gives you a translation output without touching each channel.
- Keep processing light. Channel EQ and dynamics normally carry into the send, so a high-pass filter around 80 to 100 Hz and gentle compression help. A hard noise gate that chops the start of words does not.
- Patch the bus to a physical output, or to USB channels 1 and 2 if you use the desk's built-in interface.
Pre-fader or post-fader? A post-fader send follows the engineer's fader, so when the preacher's mic is pulled down between segments, the translation hears nothing and side conversations stay out of the captions. A pre-fader send holds a steady level whatever the room mix is doing, but a forgotten open lapel mic will feed chatter straight into the translation. Many churches choose post-fader for that reason. Whichever you pick, write it on the desk so the next engineer doesn't quietly change it.
Connecting the USB interface and the computer
- Interface: any class-compliant two-channel USB audio interface with a line input will do. Translation needs one clean mono channel, so input 1 is enough.
- Line, not mic: an aux output is line level. Use the interface's line input, or set its combo jack to line. If the only option is a mic preamp, keep its gain near minimum or it will clip.
- Select the input: choose the interface as the audio input in the console and allow microphone access when the browser asks. Then check that the operating system hasn't turned the input level down (on a Mac, System Settings → Sound → Input; on Windows 11, Settings → System → Sound → Input).
- Power and sleep: run the laptop on mains power, turn off sleep for the service, silence notifications and mute the computer's speakers.
- Cables: label both ends, tape them down where feet go, and keep audio runs away from mains power cables.
Hum that disappears on battery is a ground loop. If the buzz stops when you unplug the laptop charger, the laptop and the mixer are fighting over ground. Use balanced cables, add a line-level isolation transformer between the aux output and the interface, or try a USB isolator. Never remove the earth pin or use a ground-lift adapter on a power plug to cure hum; that trades a noise problem for a safety one.
Mic technique and levels that help the captions
Recognition loves consistency: the same distance from mouth to mic, the same level, one voice at a time.
- Headset mics give the steadiest level because the distance never changes. They are the best single upgrade for a preacher who moves around.
- Lapel mics work clipped near the center of the chest, roughly a hand's width below the chin, away from necklaces, scarves and lanyards that rustle.
- Handhelds belong a few finger-widths from the mouth, pointed at it. Readers who hold the mic at chest height sound thin and distant.
- Pulpit goosenecks lose the speaker every time they step back or turn to the screen. A headset solves this.
- Wireless packs get fresh or fully charged batteries before every service, not when the low-battery light comes on.
For levels, watch both the interface meter and the console's input meter. A common target is normal speech peaking around -12 dBFS, with the loudest moments staying clear of 0 dBFS and never touching red. Too quiet and soft words go missing; too hot and consonants distort, and clipping cannot be undone afterwards. Set the level in rehearsal with the actual preacher at full sermon volume, including the loudest line of the close, because most preachers get louder at the end.
What should stay out of the translation feed?
Speech recognition tries to transcribe everything it hears. Music turns into nonsense words, reverb smears syllables together, and crowd mics create crosstalk. Keep these out of the send:
- Band and instrument channels, including the worship leader's vocal mic during songs.
- Music beds and keyboard pads played under prayers, the sermon close or the altar call.
- Playback and video audio, unless you deliberately want a clip captioned.
- Audience and ambience mics used to give the stream a sense of the room.
- Reverb and delay returns.
- The translation itself: never route the operator's computer output, the listener page or the AI voice into a channel that feeds the send.
If a music bed under the sermon close is part of your culture, keep it out of the send and accept that the room hears it while the captions don't. Songs and music don't count toward translation hours, but any music that reaches the send still turns into garbled captions, so keep it out and pause during worship as well.
A five-minute line check before the doors open
- Put headphones on the interface, or solo the send at the desk, and listen to the translation feed alone. You should hear speech mics only, with no band and no reverb tail.
- Have the preacher or a stand-in speak at sermon volume, then at their loudest, then walk the platform while talking.
- Watch the console's input meter and a phone on the listener page at the same time.
- Read one test line with a name and a scripture reference, and check how the captions handle it.
- Ask the band to play a few bars while the console is live, to confirm music isn't leaking into the send, then pause.
- Switch the console to the backup input and back, so the operator knows exactly where that control is.
This checks the audio. To judge translation quality itself, with a bilingual reviewer and an error log, use our church translation quality checklist. For the whole Sunday timeline from soundcheck to benediction, see how to translate a sermon in real time.
Common faults and how to fix them
| Symptom | Likely cause | Fix |
|---|---|---|
| Echo, doubled or smeared words | Ambience mic or reverb return in the send; a laptop mic hearing the PA | Strip the send back to dry speech channels; move from the laptop mic to a mixer feed |
| Steady hum or buzz | Ground loop between the laptop charger and the mixer, or an unbalanced cable | Test on battery; use balanced cables and an isolation transformer or USB isolator |
| Distortion on loud passages | Clipping at the interface, often line level into a mic preamp with the gain up | Use the line input or lower the gain; leave headroom for the close |
| A howl of feedback in the room | A phone or laptop playing listener audio or the AI voice near a live mic | Mute the operator computer; ask guests to use earphones for the AI voice |
| Captions stop mid-sermon | Dead wireless battery, a desk mute on a post-fader send, a sleeping laptop or a loose USB cable | Check the input meter first, then batteries, mutes, sleep settings and cables |
| Song lyrics or crowd noise in captions | Music or ambience channels in the send, or a missed pause | Rebuild the send with speech mics only; agree a pause cue |
| Quiet captions with missing words | Send level too low, or the system input level turned down | Raise the send; check the operating system's input level |
If you would like a second pair of ears on your setup, book a guided walkthrough and bring your mixer model and signal flow. When the feed sounds right on headphones, start free with the 4 translation hours every new account receives and run a full rehearsal.
Questions
Should the translation feed be pre-fader or post-fader?
Many churches use a post-fader send for live translation, so that when the engineer pulls the preacher's mic down, side conversations stay out of the captions. A pre-fader send gives a steadier level regardless of the room mix, but passes along everything an open mic hears. Choose one deliberately and label it on the desk.
Can we use the livestream mix for live translation?
On the Advanced and Premium plans, SermonLingo accepts audio from an encoder such as OBS or vMix over RTMP. Stream mixes often include music and ambience mics that make recognition harder, though. If captions struggle, build a separate speech-only send from the mixer into a USB audio interface on the operator's computer.
What USB audio interface does a church need for live translation?
A basic class-compliant USB audio interface with one line-level input is enough for live translation, because the feed is a single mono speech channel. Plug the mixer's aux or matrix output into the line input, keep the gain low enough to avoid clipping, and select the interface as the audio input in the console.
Why do translated captions show song lyrics or crowd noise?
Translated captions show song lyrics or crowd noise when music channels, ambience mics or effects returns are in the translation feed, or when the operator forgot to pause during worship. Rebuild the send with speech microphones only, and agree a clear cue with the worship leader so captions pause before the band begins.






