Most sermons are heard once and then exist only as an audio file few people replay. A transcript gives the same sermon several further uses. The church office turns it into sermon notes for the weekly email or an article for the website. Small group leaders pull the main points and quotations for discussion questions. The preacher gets an accurate record of what was actually said, which is rarely identical to the notes in the pulpit.
It is also a matter of access. Members who are deaf or hard of hearing can read the sermon in the week after the service, and so can anyone who was ill, travelling or on the children's rota. This is a transcript made from a recording after the event, not live captioning during the service.
Over time the texts become an archive that can be searched. Finding the sermon in which a particular passage or topic was preached is a text search, not an afternoon of listening.
The recording taken from the sound desk is the one to use if you have it. The preacher's microphone is fed straight into the recorder, so the voice is close and dry, and that is the condition in which a speech model makes fewest mistakes. Its one drawback is that it only contains what went through the microphones: a question from the congregation or a response from the pews may be faint or missing.
A phone recording works too, with more care. Church buildings echo, and a phone ten rows back records the room as much as the voice. Put it as near the front as you reasonably can, pointing towards the loudspeaker or the pulpit, on something soft rather than a wooden bench that carries every knock. Expect a few more errors than from a desk recording, mostly in quiet passages and when the preacher turns away.
If the service was streamed on YouTube, you do not need to download anything. Once the stream has ended and the recording is available as a public or unlisted video, paste its link in the YouTube tab. Streams on other platforms are not fetched from a link; download the video from your page there and upload the file.
A recording of the whole service contains much more than the sermon. Readings, prayers and notices are transcribed as normal speech. Sung worship is not: lyrics set to music come out poorly, and those stretches of the transcript are best ignored or deleted. The timestamps make it easy to find where the sermon starts and ends. A full service usually fits within the 3 hours limit for one recording, but trimming the file to the sermon saves both processing time and free hours.
A speech model writes what it hears. A reference spoken as John three sixteen may appear in exactly those words, or as John 3.16, or occasionally as a different number when the preacher says it quickly. Go through every chapter and verse before publishing, and set quoted passages against the translation your church uses, because a verse recited from memory and a verse as printed often differ by a word or two.
Less common biblical names are the other place to look: Habakkuk, Melchizedek, Thessalonians and place names can come out as the nearest familiar word. The same goes for names of members mentioned in notices and for Hebrew, Greek or Latin terms. A free account includes an editor, so these corrections are made once in the text and kept.
The sermon is transcribed in the language it was preached in, detected automatically from more than fifty. For congregations that worship in more than one language, translation of a finished transcript is available on the paid plans, along with AI summaries, subtitle export in SRT, VTT and ASS for the video of the service, and export to PDF or Word. A sermon with a live interpreter speaking after each sentence is transcribed in the dominant language, and the other comes out less reliably.
The free allowance is 10 recordings and 5 hours in total, shared between guest use and a free account. As a guest you see a preview, about the first ten minutes of a full-length sermon; the complete text is kept in a free account. Recognition runs on Transcrify's own servers, the audio is not sent to a third-party speech service, and audio uploaded as a guest is deleted after a day while the transcript stays.