"Will our internet handle it?" is the first question most technical volunteers ask, and it deserves a proper answer rather than reassurance. The honest version is that the two halves of the problem are completely different: the speaker's side is very light, and the congregation's side depends entirely on your building.
The speaker's connection
Sending audio out to be transcribed is a small, steady upload — comparable to a voice call and far below a video call. If your church already streams a service to the internet, this is a fraction of what you are doing already. Almost any modern broadband handles it without noticing.
What matters here is not raw speed but stability. A connection that drops for ten seconds costs you ten seconds of sermon, and no amount of headroom prevents that. A line that averages a good figure while dipping repeatedly is worse than a slower line that never dips.
Two things follow from that. Use a cable rather than Wi-Fi for the broadcasting device if you possibly can — a laptop on the same congested Wi-Fi as two hundred phones is a bad idea. And put the broadcasting device on a different network from the guest Wi-Fi if your setup allows it, so a busy room cannot slow down the source.
The congregation's connection
Here the arithmetic changes completely, because every attendee is a separate connection.
- Text subtitles are tiny. Genuinely negligible — a few kilobytes a minute per person. Two hundred people reading captions is not a meaningful load.
- Spoken translation is heavier, because audio has to reach each phone. Still modest per person, but two hundred people listening is a different proposition from two hundred people reading.
- Mobile data usually carries most of it. Most attendees never join church Wi-Fi at all, which quietly solves the problem in any building with decent mobile coverage.
Count access points, not megabits
This is the mistake that catches churches out. A single consumer router will struggle with a full congregation long before a fast line does. The limit that bites is the number of devices connected at once, not the speed of the connection behind them.
A typical domestic router starts misbehaving somewhere in the tens of devices. A hall of two hundred people, half of whom have phones that connect automatically because they joined once last year, is well past that — and the failure looks like slowness rather than a clean disconnection, which makes it hard to diagnose.
If people will be using your Wi-Fi rather than mobile data, this is the thing to spend money on: more access points, not a faster line.
The captive portal problem
Worth its own section because it defeats more first attempts than anything else on this page.
Many guest networks show a sign-in page before letting anyone through — accept the terms, enter an email, watch a message. On a phone, scanning a code while stuck behind one of those produces a blank page, an error, or a silent failure, and the visitor concludes the translation does not work and puts their phone away. They will not try twice.
Either remove the portal for the service, or make sure the connection page appears and completes before people try to join, or simply tell people to use mobile data. Test it yourself with a phone that has never joined your network — a phone that connected last year will sail through and tell you nothing.
Old buildings are the real enemy
Beautiful stone churches are excellent at blocking radio signal, and a metre of granite defeats both Wi-Fi and mobile data. Basements and crypts are worse.
The fix is unglamorous: test in the corners, in the balcony, in the side aisles and under the gallery, not standing next to the router. Where coverage is genuinely absent in part of the building, know that in advance so you can seat people who need translation where it works, rather than discovering it during a service.
Plan for the drop
Connections fail. What actually matters is how quickly someone notices.
We have seen a service lose the better part of a minute of translation because the stream dropped silently and nobody was watching the screen. The preacher carried on, the room carried on, and the only people who knew were the ones reading a page that had stopped moving.
Assign one person to keep the operator view visible. If the text stops scrolling while somebody is clearly still talking, they can act within seconds instead of minutes. This is a rota job, not a technical one.
A cheap insurance policy
Keep a phone with mobile data ready to use as a hotspot. It costs nothing on a normal Sunday and rescues the one Sunday the line goes down. Test it once so the person on duty knows how to switch over without reading instructions.
A checklist for your technical volunteer
- Broadcasting device on a cable, or at least on a separate network from the guests.
- Guest Wi-Fi tested from a phone that has never used it, including any sign-in page.
- Coverage checked in the four corners, the balcony and the side rooms.
- One person rostered to watch the operator screen during the service.
- A hotspot phone charged, and someone who knows how to switch to it.
Try it in your own building
The only reliable test is a real one, in the actual room, on the actual network, with the room full. InMyTongue runs in a browser with free credit on new accounts, so a trial run costs nothing but ten minutes. Set up a test service.