Sermon Transcription Guide, from Recording to Transcript

Sermon transcription makes the message searchable, readable for members who could not attend, and reusable in the bulletin, on the website, or as subtitles under the video. This guide covers the whole path: getting a clean recording in a sanctuary, transcribing it on a Mac or iPhone without paying per minute, labelling speakers for a Q&A, and exporting text and subtitles. Every route here runs on your own devices, so the recording never has to leave the church.

Choose your audio source

Transcription quality is decided in the room, not in the software. The best model in the world cannot recover words that were never captured clearly. You have two realistic sources.

Option A: a phone or recorder near the speaker

Put the phone on the pulpit or lectern, or on a small stand within about a metre of the preacher, screen locked and Do Not Disturb on. This is the zero-setup route and it works surprisingly well for a single voice in a quiet room. The weaknesses are reverberation in a large sanctuary, congregation noise, and anything said away from the pulpit (readings from the front pew, a question from the floor).

If you record with an iPhone, use a plain recording app or Voice Memos, and check that the phone is not muted by a case or a hand. On an Apple Watch, a recording app on the wrist is a discreet option for a preacher who moves around; the audio still needs to reach a phone or computer for transcription.

Option B: a feed from the mixing board

If the church already runs sound through a mixer, the cleanest recording is a line-level feed from it. Most boards have an auxiliary, monitor, or record output. Run that into a small USB audio interface connected to a Mac, or into a handheld recorder with a line input. You get exactly what the microphones pick up: the lapel or headset mic on the preacher, the lectern mic, and nothing else.

Two things to set up once:

1. Record a mono track at 44.1 or 48 kHz. Stereo adds nothing for speech and doubles the file size.
2. Set the level so the loudest moment peaks well below the top of the meter. Clipping ruins transcription far more than a slightly quiet recording does.

The board feed has one blind spot: it only contains what is miked. If you want a Q&A after the sermon in the transcript, pass a handheld microphone to the person asking, or accept that the questions will be quiet and label them by hand.

Transcribe the recording

With a clean file in hand, you have three routes on Apple hardware. All three run on the device.

Route 1: Voice Memos on iPhone

If you recorded in Voice Memos, recent versions can show a transcript for recordings in supported languages. It is plain text with no timestamps you can export as subtitles and no speaker labels, but for a single voice it gives you a readable draft in minutes. Copy the text out and continue with the cleanup section below.

Route 2: whisper.cpp on a Mac

whisper.cpp is a free, open-source tool that runs the Whisper speech model locally. Install ffmpeg and whisper.cpp with Homebrew, convert the recording to a 16 kHz mono WAV, and run the model:

brew install ffmpeg whisper-cpp
ffmpeg -i sermon.m4a -ar 16000 -ac 1 -c:a pcm_s16le sermon.wav
whisper-cli -m ggml-small.en.bin -f sermon.wav -otxt -osrt

That writes a plain-text transcript and an SRT subtitle file next to the audio. The full setup, including where to get the model file and what to do when a command name differs, is in Run Whisper locally on a Mac. Expect a 40-minute sermon to take a few minutes on an Apple Silicon Mac with a small model, longer with a large one.

whisper.cpp does not know who is speaking. For a sermon that is one voice this does not matter; for a Q&A, see the speaker section below.

Route 3: a transcription app with import

If you would rather not keep a Terminal setup working every week, an app that accepts a shared audio file and transcribes it in the background is the easier path. The general comparison of options on a Mac is in Transcribe audio to text on a Mac, and the one-step route for churches is described at the end of this guide.

Clean up the text for the bulletin or the website

Spoken sermons do not read like written ones. Before publishing, plan on a short editing pass:

1. Remove filler ("um", "you know", repeated starts) if your transcription tool did not already do it.
2. Check scripture references. Models hear "First Corinthians thirteen four" correctly most of the time but may write "1 Corinthians 13, 4" or spell out numbers. Decide on a house style and apply it.
3. Check proper names: people in the congregation, place names, theological terms. These are the most common errors in any transcript, and a quick find-and-replace after the first correction handles repeats.
4. Add headings at the natural turns of the sermon. Three or four H2s make a 5,000-word transcript readable on a phone.
5. Write a short summary for the top of the page or the bulletin. If you want to generate a first draft locally rather than by hand, Summarize transcripts with a local LLM walks through it.

Keep the original audio and the raw transcript. If someone asks about a quote later, you want to be able to check.

Speaker labels for Q&A and panels

A single preacher needs no speaker labels. A Q&A, a panel, or a service with several readers does, otherwise the transcript becomes one confusing block. Your options:

• Label by hand. Play the recording, and insert "Question:" and "Answer:" (or names) at each change. For a 15-minute Q&A this takes about as long as the Q&A itself.
• Use a diarization tool. Some open-source pipelines can split audio by speaker and align the result with a Whisper transcript. They are accurate on distinct voices with a decent recording and less so on a quiet floor question picked up by the pulpit mic. Setting one up is a project in its own right; Speaker diarization explained covers how the technology works and what limits it.
• Use an app that tells speakers apart and remembers them. Once the preacher's voice has a name, it stays named in every later recording, and only the guests need labelling.

Subtitles for the video recording

If the service is recorded on video and posted online, a subtitle file is the single most useful thing you can add: it makes the video searchable, accessible for people who are hard of hearing, and watchable with the sound off. The -osrt flag in the whisper.cpp command above already produced one. Load it into your video platform as a caption track or into your editor before export. If the timing is slightly off, or the lines are too long for the screen, Create SRT subtitles from audio on a Mac shows the fixes.

A weekly workflow that holds up

For a church doing this every Sunday, the routine that lasts is short and boring:

1. Sunday morning: test recording, then record from the board (or a phone at the pulpit as backup).
2. Sunday afternoon: transfer the file, transcribe, skim for names and references.
3. Monday: light edit, add headings, export text for the website and SRT for the video.
4. Archive the audio and the transcript together, named by date and passage.

Total time after the first week: about 30 to 45 minutes per sermon with the Terminal route, mostly editing. The setup itself is the part that tends to break when the volunteer who built it is away, which is the main reason to consider an app.

The one-step alternative

Frequently asked questions

How long does it take to transcribe a 40-minute sermon?

On an Apple Silicon Mac, a local speech model transcribes a 40-minute recording in a few minutes; the exact time depends on the model size and the machine. Editing takes longer than transcribing: allow 20 to 30 minutes for a careful pass over names, references and headings. The first week takes longer because of setup.

Should I record from the soundboard or with a phone?

The soundboard feed is cleaner and more consistent because it contains only the miked voices. A phone at the pulpit is good enough for a single speaker in a small or well-treated room and needs no setup. If you can, use the board as the main recording and a phone as a backup; a dead battery or an unmuted channel on the board happens to everyone eventually.

Do I need permission to transcribe and publish a sermon?

The preacher's words are their own, and most churches have a standing arrangement for recording and posting services. Check with the speaker and, if a Q&A is included, with the people who asked questions, since a published transcript is more permanent than a live comment. This is a courtesy and policy question rather than a technical one, so follow your church's own practice.

Will the transcript get Bible verses and names right?

Common references and well-known names are usually correct. Unusual names, local place names, and references read quickly are the most frequent errors, and every tool makes them. Plan a short check for those, and keep a list of corrections you make often so next week's edit is faster.