Which audio feed does live church translation need?

Give the translation one dedicated audio feed carrying the speech, and almost every problem people report disappears. The two ways to do that are an aux send from the desk with only the speaking microphones on it, or a wireless clip-on microphone on the preacher that touches nothing in your system. A multi-campus church provides that feed once, at the campus where the preacher is standing.

This is the question churches actually get stuck on. Not the languages, not the phones in the room: the signal. A church that has already tried something and walked away almost always walked away here, because what the tool was listening to was the room mix or the stream — the preacher plus a band that is soft in one minute and full in the next.

Why the room mix is the wrong feed

Your front-of-house mix is built for the room. It carries drums, keys, bass, backing vocals and the preacher, balanced for people sitting in seats. A stream mix is the same problem with compression on top.

Speech recognition on that signal is doing two jobs at once: finding the voice, then understanding it. The honest version of this — and it is true of every speech system, ours included — is that where loud full-band music sits underneath a voice, recognition degrades. Nobody has solved that, and a vendor who tells you the music does not matter is telling you something you can disprove on a Sunday.

So the fix is not a cleverer model. The fix is to stop sending the music.

Option one: an aux send with the speaking mics

On any digital console, create an aux (or a matrix, or a bus — the name varies) and assign only the speaking microphones to it: pulpit, handheld, lapel, whatever the person at the front is using. Nothing else. No band, no playback, no ambience.

That is a few minutes of configuration and it gives the cleanest result of anything you can do. It is also the version your sound team will like, because it changes nothing they can hear.

Two details that catch people out:

• Post-fader or pre-fader matters. If the aux is post-fader and someone pulls the preacher's channel down between songs, the translation goes silent with it. Pre-fader keeps the feed alive; post-fader follows the room. Pick deliberately, and tell whoever is on the desk which one you picked.
• Label the output. An unlabelled cable out of the back of a console gets repurposed within a month. Write on it.

Option two: a clip-on microphone, independent of everything

This is what we run every Sunday in our own church in Vienna. A small DJI wireless microphone attaches magnetically to the preacher's shirt and its receiver feeds the device directly. Nothing in the desk changes. Nothing in the broadcast chain changes. If the sound team is mid-service and busy, they do not need to know it is happening.

It works for a specific physical reason: the microphone sits a hand's width from the mouth. Speech arrives loud and the band arrives distant, so the music is already far quieter than the speech before any software looks at it. You are solving an audio problem with position instead of processing.

It is also the answer for a church that does not own a digital desk, meets in a school hall, or changes rooms. There is nothing to configure in a building you do not control.

Where it is weaker: you now have one more thing to charge and one more thing to hand over in the five minutes before the service. Treat it like a radio mic, because it is one — same battery discipline, same handover, same person responsible.

Multi-campus: provide it once

Churches with more than one site usually assume this gets harder. It gets easier.

The translation needs the voice where the voice is. If the preacher is at one campus and that service goes out live to the others, the feed is taken at that campus, once. Everyone listening — in that room, in the other campuses' rooms, and at home — opens the same link on their own phone and hears the translation from their own seat. Your campus-to-campus broadcast is untouched; nobody has to insert anything into it or take anything out of it.

That is worth saying plainly because the opposite assumption stops projects: we would have to rebuild our broadcast to do this. You would not. You add one send at one location and leave the rest of your infrastructure exactly as it is.

What to do this Sunday

Before you book a demo of anything, go and find out which of these two you can actually produce:

1. Ask whoever runs sound whether a spare aux output exists and whether the speaking mics can be assigned to it alone. The answer takes them thirty seconds.
2. If the answer is no, or the room is not yours, price a wireless clip-on mic instead. That is the whole workaround.
3. Then record two or three minutes of that feed on a normal Sunday — the real one, with the band doing what the band does — and keep the file. It is the only honest input for judging any tool, including ours. A quiet office test tells you nothing about your building.

That recording is also the fastest way to get a straight answer out of a vendor. Send it, and ask them to run it and show you the output. Anyone unwilling to be judged on your audio rather than their demo has told you something useful.

In the room, Sanctavo takes that one feed and carries the service on each listener's own phone from a QR code, a natural voice that keeps the speaker's tone and rhythm, under a second behind, with captions alongside. Afterwards the same recording becomes the transcript, the summary, the full message and a small-group guide, in every language the church serves — so the feed you set up once on a Sunday morning is also what produces the week's written record. If you want the comparison framework rather than the setup, what a translated service costs, Sunday by Sunday and church AI translation: what to test are the two to read next, and the equipment page covers what you no longer need to buy.