Honest expectations
Direct answer
Apple's built-in Voice Memos transcription, added with iOS 18 and macOS Sequoia, is genuinely good for clear, single-speaker speech recorded close to the microphone in a quiet room, and noticeably worse with background noise, distance, overlapping voices, heavy accents, or specialized vocabulary. It runs on-device, supports a limited set of languages, and offers no speaker labels. Treat any automatic transcript, Apple's or anyone's, as a searchable draft to check against the audio rather than a verbatim record.
Voice Memo Exporter does this in one pass: every memo into one searchable Master Transcript. $49 early access, runs privately on your Mac.
Get Early Access - $49 (Gumroad)Transcription quality is mostly decided before any software runs, by the recording itself. The same engine that nails a clear kitchen-table memo will mangle the same voice from across a room with a dishwasher running. These are the factors that matter, roughly in order.
Credit where due: for the common case of one person talking clearly into a phone, Apple's on-device transcription is solid. It requires no setup, costs nothing, processes on the device rather than a server, and shows up right in the app on iOS 18 and macOS Sequoia. For reading back a single memo or finding a phrase you know you said, it's genuinely useful, and for many recordings it will be nearly clean.
The limits are structural, not bugs. There's no batch transcription: transcripts exist per recording, viewed one at a time. There's no export button: getting text out means selecting and copying it manually. There are no speaker labels, so interviews read as one undifferentiated stream. Older recordings may need to generate on first view, and low-quality or unsupported-language recordings may produce nothing. And there's no combined view across recordings, which is the thing an archive actually needs.
These habits improve results with Apple's engine, Whisper-based tools, and everything else, because they fix the recording rather than the software.
Record closer than feels necessary
Phone within arm's length, ideally on a soft surface. Distance is the cheapest accuracy upgrade that exists.
Kill the background noise you control
TV off, windows closed, away from the road. Thirty seconds of setup saves a transcript full of guesses.
Say names and unusual terms slowly, or spell them once
Engines guess at proper nouns. Giving a name one clear, slow mention makes the rest of its appearances easier to correct.
Keep one speaker per recording when you can
If you're interviewing someone, let them finish before you talk. Overlap is where transcripts fall apart.
No consumer transcription, Apple's, Whisper's, or anyone's, is a verbatim court record. Voice Memo Exporter's local transcription is best-effort under exactly the same conditions as everything above: clear solo speech comes out well, hard audio comes out rough. What it adds is coverage, not better accuracy: transcription runs across the whole library in one pass, and the archive keeps the original audio next to every transcript, so a rough patch is always one click from the recording itself.
Voice Memo Exporter does the steps above in one pass: every Apple Voice Memo on your Mac exported, transcribed locally, and combined into one searchable Master Transcript. $49 early access, private by design.
For recordings where near-verbatim accuracy is essential, a legal matter, a quote you'll publish, a family story you want word for word, the honest workflow is automatic transcription first, then human review against the audio for the passages that matter. If a specific recording deserves the best possible automatic pass, tools with larger Whisper models are the right instrument for that one file. For everything else, a searchable best-effort transcript with the audio preserved beside it covers the real use case: finding what you said, when you said it.
Usually the recording, not the software: distance from the microphone, background noise, multiple voices, or unusual vocabulary. Re-recording closer in a quiet room fixes more transcription problems than switching tools does.
Common causes: the device isn't on iOS 18 or macOS Sequoia yet, the recording's language isn't supported, the audio quality is too poor, or an older recording simply hasn't generated its transcript yet. Opening the recording and waiting a moment often produces one.
It runs on-device, so transcription itself doesn't depend on a server connection once the feature and language support are set up on your device.
It doesn't claim to be. It's local and best-effort, subject to the same audio-quality limits. What it adds is coverage and structure: every recording transcribed in one pass, combined into one searchable Master Transcript, with the original audio preserved beside every transcript for verification.
Reliable speaker separation is mostly a cloud-service feature and still imperfect there. Most local tools, Apple's and Whisper-based ones alike, transcribe interviews as one stream, which is worth planning around if you record conversations.
Voice Memo Exporter runs this whole workflow in one pass on your Mac: every memo exported, transcribed locally, and combined into one searchable Master Transcript you own.
$49 Early Access for the first 100 buyers, then $79. One-time purchase, no subscription. 30-day refund if it doesn't work for your Mac setup.
Build a clean local archive with copied audio, transcripts, metadata, Markdown, DOCX, and PDF exports.
Get Early Access - $49 (Gumroad)$49 one-time · Forever license · No subscription · No cloud upload
