Transcription
Whenever the AI Sermon Planner needs the words of a sermon, it produces a transcript automatically. Transcripts power AI content generation, make sermons searchable, and provide accessibility for deaf and hard-of-hearing members.
How it works
Transcription runs on your device, locally, at $0 cost. We never send your audio to the cloud for transcription, and we never charge for it. When you hit Generate All Content (or any Generate button) on a sermon that doesn't have a transcript yet, the app transcribes in the background, then continues with AI generation.
When does it run?
The planner only transcribes when it needs to. It looks for a transcript in this order:
- Existing context — If you manually added substantial notes, those are used as-is
- Shared transcript from your church — If someone else in your church has already transcribed this sermon on their device, the app downloads their copy instead of re-transcribing
- Local transcription — Otherwise, the app downloads the audio, extracts it, and transcribes on-device
Once transcription finishes, the transcript is shared back to your church (encrypted — more below) so other staff members on other devices skip transcription entirely next time.
Progress indicators
While transcription runs, the planner shows a live status on the Generate button:
- "Checking for existing transcript..." — looking for a shared copy from your church
- "Downloading AI model (first time only)..." — one-time setup on first use (~75-150 MB)
- "Getting sermon audio..." — downloading the recording
- "Preparing audio..." — extracting and converting for the AI model
- "Listening to sermon..." — the actual transcription
- "Transcribing chunk X of Y..." — for long sermons, the audio is split into parallel chunks
After the first sermon, the AI model is cached — every subsequent sermon skips that download.
Desktop vs. mobile
Desktop (Windows, macOS, Linux):
- Handles any sermon length
- Uses a larger, more accurate AI model
- Splits long sermons into chunks and transcribes in parallel — significantly faster on multi-core machines
- Memory-aware: adapts to your laptop's RAM so the rest of your system stays responsive
Mobile (iOS, Android):
- Works great for sermons up to about 90 minutes
- Uses a lightweight AI model to fit in phone memory
- For sermons over 90 minutes, the app will suggest using the desktop app or adding manual notes instead
For long or frequent use, the desktop Church Admin app is recommended.
Language support
Language is auto-detected from the audio. Supported languages include:
- English, Spanish, French, German, Italian, Portuguese
- Indian languages: Hindi, Telugu, Tamil, Kannada, Malayalam, Bengali, Punjabi, Marathi, Gujarati
- 90+ others
No configuration needed — just preach, and the app figures it out.
Privacy & encryption
- Transcripts are encrypted per-church before being shared to your church's cloud storage — each church has its own unique key
- Keys are never transmitted or stored by us; they're derived from your church's ID on-device
- Nlarj staff cannot read your transcripts
- Only devices logged into the same church can decrypt the transcript
- Transcripts are stored in your own cloud storage, not ours
Accuracy tips
Transcription accuracy is mostly a function of audio quality. For the best results:
- Lavalier mic on the pastor — close-mic'd audio is near-perfect
- Close room mics — 3-4 feet from the pulpit
- Quiet room — HVAC, fans, and congregation noise degrade accuracy
- Consistent volume — use a compressor/limiter in your audio chain
Typical accuracy:
| Audio quality | Accuracy |
|---|---|
| Studio-quality mic | 95-98% |
| Lavalier mic | 92-96% |
| Handheld mic with good technique | 88-94% |
| Room mic only | 70-85% |
| Phone recording from the back | 50-70% |
Reuse across your church
Once any staff member transcribes a sermon, everyone else's devices reuse that transcript automatically. Practical upshot:
- Your media volunteer transcribes Sunday morning on a beefy laptop
- Your pastor opens the same sermon on a tablet that evening — transcript is already there, AI generation starts immediately
Troubleshooting
"No recording found for this sermon" — The sermon has no audio or video file attached. You can still use the planner by typing notes or a topic manually.
Transcription is slow — Expected on long sermons. Desktop machines with more CPU cores finish faster. First-time use also includes a one-time AI model download.
App feels sluggish during transcription — The app dynamically adapts worker count to your RAM and CPU to stay responsive, but heavy use (another app is recording, rendering video, etc.) can slow it down. Try closing other heavy apps.
"Sermon audio is too long for mobile transcription" — Sermons over 90 minutes need to be transcribed on the desktop app. Alternatively, type manual notes to drive AI generation without a full transcript.
Wrong language detected — Usually an audio quality issue. Try a better mic or adding manual notes so the AI has clear context.