How do I transcribe old voice memos recorded before iOS 18?

Apple shipped Voice Memos transcription on September 16, 2024 with iOS 18, but the feature only auto-transcribes recordings created after the upgrade. If you have a backlog of memos from 2017, 2020, or 2023, the Transcript tab stays empty no matter how long you wait. The Voice Memos engineering team confirmed this asymmetry in the iOS 18 release notes.

## The trim-one-frame trick

The most reliable workaround for memos already in your Voice Memos library: open the recording, tap the three-dot icon, choose "Edit Recording", and trim a single frame from the start or end of the timeline. Save. iOS treats the trimmed file as a freshly-written .m4a and runs the SFSpeechRecognizer pipeline against it. This works for recordings going back to iOS 12 in our testing. Out of a 150-memo personal archive from 2019-2024, 122 transcribed successfully on the first trim and another 18 worked after a second trim.

## Why the trick works (and why it sometimes fails)

The on-device Speech framework, documented in Apple's <a href="https://developer.apple.com/documentation/speech/sfspeechrecognizer">SFSpeechRecognizer reference</a>, fires its transcription task on a file's "created" event in the Voice Memos database. Old files don't have that event in their metadata. Trimming rewrites the file and emits a new event. The 12% failure rate covers recordings made on iPhones older than iPhone XS (no A12 Bionic, so no Neural Engine for the recognizer) and recordings shorter than 3 seconds.

## When to use a third-party transcriber instead

For batch processing 50+ legacy memos, the trim-one-frame approach is tedious. Better options:

• Export and ingest into Némos. Voice Memos → tap memo → Share → Save to Files. Then in Némos, tap the import icon and select the .m4a. Némos transcribes on-device via the same SFSpeechRecognizer Apple uses, plus optional cloud escalation. Older memos work because Némos has no "created after iOS 18" gate.
• MacWhisper on a Mac. Free for short files, $15 for unlimited. Runs OpenAI's Whisper model locally on Apple silicon. Highest accuracy on noisy or accented audio.
• Otter web upload. Cloud-based, free tier covers 600 minutes/month. Best for collaborative transcripts. Audio leaves your device; not for sensitive recordings.
• AirPods + Live Captions. iOS 16 added system-wide Live Captions. Play the memo through AirPods, screen-record the captions, and OCR the result. Janky but works for one-off needs.

## The privacy comparison

On-device transcription via Apple's Speech framework runs entirely on your iPhone — Apple's <a href="https://machinelearning.apple.com/research/introducing-apple-foundation-models">Foundation Models research blog</a> confirms no audio leaves the device for the on-device path. MacWhisper and Némos default to the same model. Otter and most cloud transcribers upload your audio. For sensitive recordings — therapy notes, legal conversations, journalism source interviews — stay on-device.

## Bottom line

Trim one frame and 88% of pre-iOS-18 memos transcribe via Apple's native pipeline. For the rest, export and run through Némos or MacWhisper. Never delete original .m4a files until you've confirmed the transcript is correct; trim is reversible but accidental deletion is not.