Live translation handles a spoken talk well. Music is a different animal entirely. If you have ever watched a live system try to keep up with a sung chorus you will know the result can be worse than no translation at all — and understanding exactly why points straight at the fix.
Why songs break live translation
Speech recognition is built for speech: clear consonants, natural pauses, one voice, a steady pace. A sung line has none of those properties.
- Melody stretches vowels and softens the consonants that carry most of the information in a word.
- The band fills the space. A speech engine has to separate a voice from a full mix, which is a much harder problem than separating it from a quiet room.
- The congregation sings along, so the microphone hears two hundred people at slightly different pitches and a fraction of a second apart.
- Words repeat and overlap. A chorus sung four times produces four attempts at the same line, none of them agreeing.
- Lyrics are poetry. Even a perfect transcript would translate badly, because compressed, metaphorical language is exactly what literal translation handles worst.
It is not really a flaw in the technology. It is a mismatch between a tool and a task, and no amount of tuning fixes it.
And it fails in the worst way
The problem is not that the text is missing. It is that the text is wrong and looks confident. A congregation reading a garbled, half-invented version of a worship song is having a worse experience than one reading nothing, because they are trying to make sense of something that does not make sense.
For a sermon, an occasional error is tolerable. For a song people are trying to sing, it is not.
The approach that works: prepared lyrics
The reliable answer is to translate the songs before the service rather than during it. With the lyrics for today's songs prepared in advance, the system displays the correct translation line by line as the song plays. No live guessing, no scrambled text, and a congregation that can actually follow — or sing.
Why prepared translation is better anyway
- Accuracy. A song translated ahead of time can be checked by somebody who speaks the language. A live one cannot be checked by anybody.
- Singability. Worship lyrics carry rhythm and stress as well as meaning. A translation that scans can be sung; a literal one can only be read. That is a judgement only a person can make.
- Theology. Songs compress doctrine into short lines, and a loose rendering can change what a line says. This is the strongest argument for a human pass — and for the same reason it is worth having a native speaker check anything you use repeatedly.
- Reuse. Translate a song once and it is ready every time you sing it again. Most churches use a rotating set of perhaps forty songs, so the work is finite and it compounds.
Building the library is smaller than it sounds
Forty songs sounds like a project. In practice it is not, because the load is front-heavy and then almost nothing.
- List what you actually sing. Most churches discover the real list is thirty to fifty songs, not the hundreds in the songbook.
- Start with the ten you sing most. These will cover the majority of any given service.
- Do the ones with the most repeats first, since a chorus sung four times is four times the value.
- Get a native speaker to read each one before it goes into use. Half an hour with someone from the congregation is worth more than any automatic check.
- Add new songs as you learn them, which is two or three a term rather than a backlog.
Ask a member who speaks the language to help. In our experience people find this genuinely enjoyable in a way that most volunteer jobs are not — it is creative work, it is theirs, and the result is visible on a screen every Sunday.
Copyright, briefly
Worth checking rather than assuming. Most licences that cover displaying lyrics also cover translating them for display, but the terms vary and some require you to note the translation. It takes one email to your licensing body and it is better asked now than after somebody notices.
Keep speech and song separate
The practical rule is to let each tool do what it is good at: live translation for the sermon, the prayers, the notices and the readings; prepared lyrics for the music. Some systems will switch between the two automatically as the service moves.
Blending them this way gives a congregation the best of both, and spares everyone the experience of watching a machine wrestle a melody and lose.
The parts that are neither speech nor song
A worship set is not only songs. Between them sit the bits that are hardest to plan for, and it is worth deciding in advance what happens during them.
- Spoken prayer over music. This is speech and it translates well, but the band underneath makes it harder. If someone regularly prays over a pad, ask them to use a microphone properly rather than holding it low.
- Spontaneous singing. Nothing prepared exists and nothing sensible can be produced live. The honest answer is to show nothing for those minutes rather than show nonsense.
- A song in a language most people do not speak. Increasingly common in multilingual congregations, and a good use of prepared lyrics in reverse: show everyone else what is being sung.
- The bridge into the sermon. Make sure the handover is clean, so the first minute of the talk is not lost while somebody switches modes.
None of these needs solving perfectly. They need deciding, once, so that the person operating on a Sunday is not making a judgement call in the middle of a service.
Sing it, do not only read it
One last thing worth aiming at. There is a real difference between a congregation reading a translation of a song and a congregation singing it.
If the translated lines scan — if they fit the tune well enough to be sung — people will sing them, quietly at first and then not quietly. A room where three languages are singing the same line at the same time is a different thing from a room where most people are watching. It is worth the extra effort of asking whoever translates to sing each line themselves before signing it off.
Try it at your next service
InMyTongue keeps a library of pre-translated song lyrics alongside live translation of the spoken parts, so the music is accurate and the sermon is live. New accounts include free credit. See how it works.