Transcribing Spanish and bilingual sermons, faithfully in both languages.

If your church preaches in Spanish, in English, or, like many Hispanic congregations in the US, in both at once, you’ve probably found that most transcription tools weren’t built with you in mind. Spanish sermon transcription is usually an afterthought, and a bilingual sermon is where it shows: the moment the preacher says “Dios es bueno, ¿amén? He is good,” the transcript falls apart.

This page is about that moment. We’ll explain why switching languages mid-sermon is genuinely hard for speech recognition, what has changed, and how SermonKeep handles Spanish and bilingual preaching today, including what we haven’t tested yet. Full disclosure: SermonKeep is our product.

The sermon that lives in two languages

If you’ve spent time in Hispanic churches in the US, you know the pattern. The pastor opens in Spanish for the abuelas in the second row. A key point lands in English for the teenagers who think in it. The Scripture is read from the Reina-Valera and applied with a verse from the NIV. The whole room says “amén,” and nobody finds any of this remarkable, because this is how a bilingual community talks.

Linguists call it code-switching: moving between languages fluidly, often inside one sentence, choosing whichever carries the thought best. It isn’t sloppy speech. It’s how a bilingual preacher reaches a bilingual congregation, each generation in the language it prays in.

It also means a bilingual sermon isn’t “a Spanish sermon” or “an English sermon.” It’s both, interleaved, and a transcript that serves the church has to be faithful to both.

Why transcription used to fall apart at the switch

A sermon preached entirely in Spanish is, honestly, a solved problem: modern speech recognition handles Spanish audio well. If that’s your church, you need the same things an English-speaking church needs: a decent recording and a service that knows sermon vocabulary. Our explainer on how sermon transcription works walks through the whole process.

Mixed sermons were the hard case, for one structural reason: most transcription tools locked the whole recording to a single language. Every word the model was allowed to write came from that language, so when the preacher switched, the model forced the sounds into the language it expected:

• Locked to Spanish, “He is worthy” came out as Spanish-shaped nonsense.
• Locked to English, “gracia sobre gracia” came out as “grass, yeah, so bright.”

Those weren’t small errors. Whole passages, often the ones the preacher switched languages because they mattered, came out as gibberish.

What has changed

The newest speech recognition models are multilingual: one model that recognizes which language is being spoken as it goes, and writes each phrase in the language it heard. AssemblyAI, whose Universal-3.5 Pro model handles SermonKeep’s transcription, lists code-switching as supported across its languages, with Spanish-English (“Spanglish”) among them, per its model reference (October 2026).

In practice, in SermonKeep:

• You can pick the sermon’s language (English, Spanish, Portuguese, French or German) or let it be detected automatically.
• For a bilingual sermon, pick the language most of it is in. The model follows the preacher across the switch either way; the choice decides which theological dictionary we send (more on that below). Auto-detect works too, and uses the English dictionary.
• Nothing gets translated. Spanish stays Spanish and English stays English, exactly as preached.

One honest caveat: we haven’t run a formal test on bilingual sermons yet, and AssemblyAI’s published accuracy figures are measured on general audio, not on bilingual preaching in a reverberant sanctuary. Try it on one of your own sermons before you rely on it; the first one is free.

The vocabulary problem, in two languages

Every church has a vocabulary problem: biblical names and theological words that general models rarely hear. We wrote a full page on the theological dictionary we built for it. In a bilingual church the problem doubles, and the two halves don’t line up neatly:

• Names change between languages. Melchizedek is Melquisedec, Nebuchadnezzar is Nabucodonosor, Zacchaeus is Zaqueo, Gethsemane is Getsemaní.
• Doctrinal terms are near-twins. Propiciación and propitiation, santificación and sanctification: close enough to blur, different enough to matter.
• Hebrew and Greek words are spelled differently. Jésed in Spanish, hesed in English; Yahvé and Yahweh; ruaj and ruach.
• Two Bible translations are in play. A Reina-Valera quotation has its own centuries-old cadence, just as the King James does in English.

That’s why SermonKeep keeps two dictionaries. When you choose Spanish, we send a Spanish description of what a sermon sounds like (mentioning Reina-Valera and NVI quotations) and a list of about 130 Spanish terms: Bible books, Old and New Testament names and places, doctrine, and Hebrew and Greek words in their Spanish spelling. English sermons get their own list of about 180.

Honest caveats for bilingual churches

We’d rather earn your trust than overclaim, so here’s where we’d still keep a person in the loop:

• Check names in both languages. Your members, your neighborhood, your sister church abroad: no model has heard them before.
• Rapid word-by-word switching is the hardest case. A preacher who flips every few words (“Porque God tiene un plan, hermanos”) is harder than one who switches by the sentence. Expect the occasional blurred boundary.
• Community words get approximated. Words a congregation coined, like parquear or el lonche, may come back as the nearest dictionary word in either language.
• The dictionary follows the language you pick. In a half-and-half sermon, the biblical names in the other language don’t get the dictionary’s help, so they deserve a quick look.

After transcription, the review pass helps here too: obvious misrecognitions are fixed and listed so you can undo them, and anything uncertain waits for you, with the audio ready to play.

A transcript the whole congregation can hold

For a bilingual church, a faithful transcript does something quietly important: every generation can hold the same sermon. The English-dominant grandson can find it; the Spanish-dominant grandmother can reread her pastor’s words exactly as they were said. Nothing was flattened into one language to suit a piece of software.

If you’d like to transcribe a Spanish sermon, or the mixed kind your church actually preaches, our sermon transcription service does your first sermon free, up to an hour, with no card. Preach the way your congregation talks; the transcript should keep up.

Sources

• AssemblyAI, Models reference: Universal-3.5 Pro languages and code-switching support (accessed October 2026).