iPhone Voice Memo Transcription: the iOS 18 guide (and what to do when it isn't enough)

TL;DR — which path is yours

iOS 18 built-in transcription — full walkthrough

Apple shipped on-device Voice Memos transcription in iOS 18 (September 2024). It runs entirely on the iPhone’s Neural Engine, works offline, and produces a searchable transcript you can copy out. For most one-person voice memos on a recent iPhone, this is the right answer and you should stop reading this page.

Step-by-step

1. Update to iOS 18— Settings → General → Software Update. If your iPhone is 12 or newer, you’ll see the option.
2. Open Voice Memos— it lives in the Utilities folder by default, or Spotlight-search “Voice Memos.”
3. Record a new memo (or tap an existing one that was recorded after the iOS 18 install).
4. Open the transcript view — tap the recording to expand it, then tap the three-dot menu (…) at the top-right and choose View Transcript. Older iOS 18 point releases showed a small “Aa” icon; newer releases put it in the menu.
5. Copy or share — either long-press text in the transcript to select and Copy, or use Copy Transcript from the menu to grab the whole thing.

What you get

• A single running text block with punctuation and automatic paragraphing on natural pauses.
• Text is searchable within Voice Memos — swipe down for the search bar and type any word.
• The transcript view links back to the audio: tap a word to seek playback to that moment.

What you don’t get

• No speaker labels. The transcript is one running block regardless of how many people talked.
• No file export. You can only copy the text — no .srt, no .vtt, no .docx, no .json.
• No timestamps in the copied text. The in-app view lets you tap-to-seek, but the copied text is plain — no timecodes.
• No in-place editing.Mis-heard word? Copy-paste elsewhere to fix; you can’t edit the transcript inside Voice Memos.

Device and language requirements

Supported iPhones

Apple’s on-device transcription requires the A14 Bionic chip or later, which means:

No software update will add the feature to unsupported models. The Neural Engine on the A13 and earlier doesn’t have the capacity for Apple’s on-device speech model.

Supported languages (iOS 18)

Ten languages are covered at launch, with regional variants handled where relevant:

• English (US, UK, Australia, Canada, India, Ireland, New Zealand, Singapore, South Africa)
• Spanish (Spain, Mexico, US)
• Portuguese (Brazil)
• French (France, Canada)
• German
• Italian
• Japanese
• Korean
• Mandarin (Simplified Chinese)
• Cantonese (Traditional Chinese)

Not supported: Turkish, Arabic, Russian, Hindi, Bengali, Urdu, Dutch, Polish, Vietnamese, Thai, Indonesian, Swedish, Norwegian, Danish, Finnish, Greek, Hebrew, and most others. If you record in one of these, the built-in transcript is either empty or a Latin-script approximation of the sounds (rarely useful). For unsupported languages, use the upload path below.

When the built-in feature isn’t enough

The clean cases where iOS 18 works are narrower than Apple’s marketing suggests. Here’s a checklist of when to skip the built-in path:

• iPhone 11 or earlier — no on-device option, period.
• Recording predates your iOS 18 install— Apple only transcribes new recordings. Existing memos don’t backfill after the update. This is the #1 complaint in Apple Discussions threads and there’s no official workaround.
• iPhone SE (any generation)— despite the SE 3rd gen having the A15 chip, Apple restricts the feature to the numbered iPhone 12+ line. The SE doesn’t get it.
• Language not on the 10-language list. Turkish, Arabic, Hindi, Russian, and dozens of others produce empty or garbage output.
• Two or more speakers.The transcript is one undifferentiated block. If you recorded an interview or a meeting as a memo, you need diarization — the built-in doesn’t have it.
• You need .srt for a video edit. Voice Memos only exposes copy-to-clipboard. No subtitle file, no timestamped export.
• You need to edit the transcript in place (fix mis-heard proper nouns without leaving the tool). The built-in view is read-only.
• You have 50+ old memos to process.Even if they were made after iOS 18, transcription happens per-recording and there’s no batch view. Export as a bulk share and upload to a service that handles batches.
• Recording is very long (> 60 minutes). On-device transcription of multi-hour files can pause, fail to complete, or drain the battery visibly on 12/13 mini devices.
• You want higher accuracy on technical vocabulary. Apple’s on-device model is roughly Whisper-small class — solid on general speech, weaker on medical, legal, and engineering jargon. Cloud services running Whisper large-v3 or a commercial ASR are meaningfully more accurate on those.

Transcribing Voice Memos on Mac (Sequoia + Apple Silicon)

macOS Sequoia (14.7 and later) added the same on-device transcription to the Mac Voice Memos app. Requirements:

• Mac with Apple Silicon (M1, M2, M3, M4, or later). Intel Macs are not supported — the on-device speech model requires the Apple Neural Engine.
• macOS Sequoia 14.7 or later.
• Same 10 supported languages as iPhone.

How Voice Memos syncs iPhone → Mac

1. On iPhone: Settings → [your name] → iCloud → “Show All” → Voice Memos → toggle Sync this iPhone on.
2. On Mac: System Settings → Apple Account → iCloud → Voice Memos → enable.
3. A recording made on iPhone appears in Voice Memos on Mac within a minute or two (assuming both devices are online).

Viewing the transcript on Mac

Right-click a recording in the sidebar and choose View Transcript. The transcript pane opens on the right. On Mac you can select text with a normal drag and copy with ⌘C — noticeably easier than the iPhone long-press selection.

Why the Mac path matters even if you have a new iPhone

If you want to pipe the transcript into another tool (Notion, Notes, Obsidian, a script), copying from Mac is faster than fighting the iPhone selection UI. And if you’re transcribing a batch of memos, the Mac layout lets you jump between recordings without losing your place.

The upload path — exporting .m4a and using a service

For every case the built-in feature doesn’t cover, the workflow is the same: export the recording as .m4a and upload it to a transcription service.

Step 1 — Export the recording

Three ways off the phone:

1. Share menu (fastest). Voice Memos → tap recording → three-dot menu → Share → choose Save to Files, AirDrop, Mail, or any installed app. The file is .m4a(AAC audio in an MP4 container). If you’ve ever used the memo as a ringtone, it may save as .m4r — rename to .m4a if a transcription service refuses it.
2. iCloud Drive. With Voice Memos iCloud sync on (see Mac section above), your recordings live in iCloud Drive → Voice Memos folder, reachable from any Mac or from icloud.com.
3. USB-C / Lightning cable.Plug the iPhone into a Mac and open the Mac’s Voice Memos app — recordings sync via Finder on recent macOS versions. On Windows, use the Apple Devices app.

Step 2 — Upload to a service

Any service that accepts .m4a works. Concrete step-by-step with DeluxeScribe:

1. Sign up (60 minutes free, no credit card required).
2. Drag your .m4a into the upload area, or click to browse.
3. Language auto-detects — leave it unless you want to force a specific dialect. Speaker labels default on; toggle off for solo memos.
4. Wait 1–3 minutes for a typical short memo; longer recordings scale roughly linearly.
5. Review in the browser editor, fix any mis-heard proper nouns or numbers, then export as TXT, DOCX, PDF, SRT, VTT, or JSON.

Built-in vs upload — feature comparison

The honest way to think about this: iOS 18’s built-in is free and private for the narrow slice it covers. Cloud services handle everything else.

Positioning: if the iOS 18 built-in fits your case, use it — it’s free and private. If it doesn’t (older iPhone, unsupported language, need speaker labels, need .srt), DeluxeScribe is our honest recommendation because we support the widest language set at the lowest per-minute price, but Rev and Otter are also valid picks depending on your workflow.

Speaker labels, timestamps, and .srt export

These three features are the most common reasons people move from the built-in to a service. Quick primer on each:

Speaker labels (diarization)

The model separates the audio by voice and tags each segment “Speaker 1,” “Speaker 2,” etc. — you rename them post-hoc. Essential for interviews, meetings, and any multi-person memo. Diarization is genuinely hard: even the best models make attribution errors on overlapping speech, on phone-quality audio, and when two speakers have similar voices. Realistic diarization accuracy on clean voice memo audio is 80–90% correct attribution; on noisy or overlapping speech, expect 60–75%.

Word-level timestamps

Each word (or token) is tagged with its start and end time in milliseconds. Used for: podcast show notes with “click-to-jump” UIs, video editors that align transcript edits to audio cuts, court reporting, and anywhere you need to prove “this word was said at this second.” DeluxeScribe includes word-level timestamps in JSON export by default.

.srt and .vtt subtitle export

.srt and .vttare the standard subtitle formats for video. If your voice memo will become a video’s soundtrack or subtitle track, you need one of these — the built-in Voice Memos transcript can’t produce them. Every serious transcription service offers both; our SRT generator walks through the format if you want the details.

Privacy — on-device vs cloud, honestly

This is the section vendor blogs skip because it doesn’t favor the vendor. The honest read:

iOS 18 built-in — actually on-device

Apple’s speech model runs on the iPhone’s Neural Engine with no network call. The audio does not leave the device. The transcript does not leave the device unless you copy or share it. This is verified behavior — the feature works in Airplane Mode. For genuinely sensitive content (medical intake notes, legal client conversations, confidential business), this is the correct default.

Cloud services — they need the audio

DeluxeScribe, Rev, Otter, HappyScribe, and every other SaaS transcription tool necessarily upload your audio to their servers to run the transcription. Reputable services encrypt in transit and at rest, and offer data-deletion controls. DeluxeScribe deletes uploaded audio after 30 days by default (configurable) and never uses customer audio to train models. But it is upload — a fundamentally different trust model from on-device.

Self-hosted Whisper — the third option

If you need on-device processing but the built-in doesn’t fit (unsupported language, older device, need .srt), self-hosted Whisper is the private fallback. Install via pip install openai-whisper or Homebrew’s whisper-cpp, run against your .m4a, get a transcript with no upload. Cost: some technical setup and a few minutes per hour of audio on a modern Mac.

The HIPAA question

Apple’s on-device transcription is not HIPAA-certified — Apple doesn’t sign BAAs for consumer features — but because no data leaves the device, many compliance teams treat it as equivalent to writing longhand notes. Cloud services that offer HIPAA BAAs (Rev, some tiers of DeluxeScribe) are the compliant path for regulated healthcare use; see our medical transcription guide for the full framework.