Apple Intelligence Transcription: What It Does, Where It Stops

Apple Intelligence transcription is real, it’s built into macOS 26 and iOS 26, and it’s better than most people expect — Notes can record and transcribe audio, the Phone app can record and transcribe calls, and Voice Memos shows a transcript of every recording. If you’re searching this phrase, you’re probably asking one of two questions: what can Apple’s built-in AI actually transcribe for me, and is it enough, or do I still need an app?

The short answer: the built-ins are genuinely useful for casual capture, and they run on the same on-device speech stack that makes a Mac the best transcription machine Apple has ever shipped. But they stop well short of a working transcription workflow — they can’t capture a meeting’s far-end audio, they produce raw text rather than structured notes, and their export story is “select all, copy.” This post lays out exactly what Apple Intelligence transcribes natively, how the underlying stack works, and where a dedicated app takes over.

What Apple Intelligence transcription covers out of the box

Three system apps transcribe audio natively on current Apple OS versions, each scoped to its own corner:

Notes. You can record audio directly inside a note, and Notes produces a transcript alongside the recording. With Apple Intelligence enabled, it can also summarize that transcript. This is the closest thing to a built-in “recorder plus notes” flow, and for a quick in-person conversation it works.

Phone and FaceTime. The Phone app can record a call — it announces the recording to everyone on the line first — and saves the recording with a transcript into Notes. Useful for the occasional call you need on the record, with the announcement making consent explicit.

Voice Memos. Every memo gets a transcript you can view, search, and copy. If you’ve been recording ideas on your iPhone for years, that back catalog became searchable text — the workflow for turning those into filed notes is covered in the Voice Memos to Markdown guide.

All three share the trait that matters: the speech-to-text runs on-device, on the Neural Engine. Your audio isn’t shipped to a transcription API. That’s the architectural line that separates Apple’s approach from the cloud notetakers, and it’s the same line explored in depth in the on-device meeting transcription write-up.

The stack underneath: SpeechAnalyzer and Foundation Models

“Apple Intelligence transcription” isn’t one monolithic feature — it’s two frameworks that Apple also exposes to third-party apps, which is why the story doesn’t end with the built-in apps.

SpeechAnalyzer is the on-device speech-recognition framework Apple introduced at WWDC 2025. It replaces the older SFSpeechRecognizer pipeline with a model built for long-form audio — meetings, lectures, interviews — running entirely on the Neural Engine. Per Apple’s benchmarks it transcribes roughly 55% faster than Whisper v3 Turbo on the same chip. This is the engine behind the Notes and Voice Memos transcripts, and it’s the reason an hour of audio transcribes in minutes on an M-series Mac with no upload.

Foundation Models is Apple’s framework exposing an on-device large language model — the piece that turns a transcript into a summary, key points, and action items. The built-in apps use it for Notes summaries; third-party apps can call it directly for more structured output.

The significance for anyone shopping for a transcription tool: the hard parts — a fast local speech model and a local LLM — now ship with the OS. The difference between tools is no longer whose cloud transcribes better; it’s who builds the best workflow on top of the local stack. One nuance worth knowing: Apple Intelligence can route some heavier requests to Private Cloud Compute, Apple’s server tier. The transcription itself stays on-device, but if your requirement is “nothing ever leaves the machine, full stop,” that’s a distinction to keep in view — and one a fully local app can make unconditional.

Where the built-ins stop

Use Notes, Phone, and Voice Memos for what they are: capture apps with transcripts bolted on. Five gaps show up as soon as you try to run real work through them.

1. No meeting capture

This is the big one. Notes and Voice Memos record the microphone. In a Zoom, Teams, Meet, or Webex call, the other participants’ voices come out of your Mac’s speakers as system audio — your mic never hears them cleanly, and if you’re wearing headphones it doesn’t hear them at all. Record a video call with Notes and you get a transcript of one side of the conversation. The Phone app handles only phone and FaceTime calls, with an announcement, not your Tuesday standup.

Capturing the far end requires recording system audio scoped to the meeting app — a ScreenCaptureKit capability that no built-in transcription app uses. The mechanics are in the system audio recording guide; the practical upshot is that Apple Intelligence transcription, as shipped in the system apps, cannot transcribe your meetings.

2. Raw transcripts, not notes

The built-in output is a wall of text. Notes can summarize it, but the summary is a paragraph — not decisions, action items, and owners, and nothing links a summary line back to the audio it came from. When a summary bullet looks wrong, your recourse is re-reading the transcript or re-listening from the start. A purpose-built tool anchors every summary point to the exact moment in the recording, so verifying a claim takes two seconds instead of a scrub through an hour of audio.

3. Export is an afterthought

There’s no “export as Markdown to Obsidian,” no SRT, no structured file output. You select text and copy it. If your transcripts need to become something — notes in a vault, a document, subtitles — the built-ins hand you raw material and leave the pipeline to you.

4. Your transcripts scatter across three apps

Call transcripts land in Notes. Memo transcripts live in Voice Memos. In-note recordings live in whichever note you made. There’s no single searchable library of everything you’ve recorded and transcribed, which is exactly what you want the day you need to find “that conversation about pricing from March.”

5. No workflow for volume

Transcribing one memo is fine. Transcribing a weekly load of meetings, interviews, and lectures needs batch handling, consistent summaries, and organization by default. The system apps aren’t built for that cadence, and it shows quickly.

What to use when the built-ins run out

The good news hiding in the framework story: an app built on SpeechAnalyzer and Foundation Models keeps every property that makes Apple Intelligence transcription attractive — on-device processing, Neural Engine speed, no audio upload — while fixing the workflow gaps. That’s precisely what Dictanta is: the same local stack the system apps use, wired into a complete pipeline.

Concretely, on a Mac running macOS 26:

1. It captures meetings. Dictanta records your mic and the system audio of the meeting app via ScreenCaptureKit — both sides of a Zoom call, a Teams call, or any browser-based meeting, with no bot joining and nothing streamed anywhere. Press ⇧⌘R when the call starts.
2. It transcribes with SpeechAnalyzer. Same engine as the built-ins, pointed at meeting-grade audio, running on the Neural Engine minutes after the call ends.
3. It summarizes with Foundation Models — anchored. The on-device LLM produces decisions, action items, and key points, and every bullet links to the exact moment in the audio it came from. Click a line, hear the moment, confirm it’s right.
4. It exports. Markdown for Obsidian, Bear, or Notion, plus other structured formats — the transcript becomes an artifact instead of a copy-paste job.
5. It keeps everything in one library. Meetings, imported Voice Memos, and dictated notes in one searchable place, on your SSD, with processing that never requires a network — the same property that makes offline transcription on a Mac work on a plane.

The free tier covers your first three recordings with no length cap, which is enough to run a real meeting through it and watch Activity Monitor confirm nothing is uploading.

What you need to run any of this

Apple Intelligence transcription — built-in or app-based — has hardware requirements, because the models run on the Neural Engine:

• Mac: Apple Silicon (any M-series), macOS 26 Tahoe. System-audio meeting capture is Mac-only.
• iPhone: A17 Pro or newer (iPhone 15 Pro onward), iOS 26.
• iPad: M-series iPad, iPadOS 26.

Intel Macs are out — the local models require Apple Silicon. If your machine qualifies for Apple Intelligence, it qualifies for everything in this post.

Built-in vs. dedicated, side by side

Capability	Notes / Voice Memos / Phone	Dictanta
On-device transcription	Yes (SpeechAnalyzer)	Yes (SpeechAnalyzer)
Meeting far-end (system) audio	No	Yes, Mac
Phone-call recording	Yes, with announcement	No
Summary	Paragraph in Notes	Decisions, action items, key points
Audio-anchored summary bullets	No	Yes
Markdown / structured export	No	Yes
Single searchable library	No — split across apps	Yes
Voice Memos import	—	Yes
Cost	Free with the OS	Free first 3 recordings, then paid

The honest read: if you occasionally record an in-person chat or want your old memos searchable, the built-ins are free and already on your machine — use them. The moment your recordings are meetings, or your transcripts need to become structured notes somewhere else, the built-ins stop and a dedicated app on the same stack starts. Privacy isn’t the tradeoff it used to be, either way — the comparison that used to be “convenient cloud vs. private local” is now local vs. local, as the private transcription app breakdown covers.

Common questions

Does Apple Intelligence transcribe Zoom or Teams meetings? No. The built-in apps record the microphone (or, for the Phone app, phone and FaceTime calls). None of them capture the system audio where a video call’s far end lives, so a meeting recorded with Notes transcribes only your side. Meeting transcription requires an app that records system audio — on a Mac, with no bot.

Can it transcribe an audio file I already have? Not directly in the system apps — the built-in transcripts attach to recordings those apps made. To transcribe an existing .m4a, .mp3, or .wav, import it into an app built on SpeechAnalyzer; Dictanta’s Voice Memos and file import runs the same on-device pipeline over audio from anywhere.

Does Apple Intelligence transcription work offline? Yes — that’s the point of the architecture. The speech model lives on the device, so transcription works in airplane mode, and a third-party app on the same stack inherits the property. Summaries in the built-in apps may want Apple Intelligence available, but nothing requires a transcription server.

Which languages does it support? SpeechAnalyzer ships with a growing set of supported locales, with English the most mature. If you transcribe primarily in another language, test it on a real recording before committing to any tool built on the stack — quality varies by locale, and the free tier exists for exactly this kind of check.

Bottom line

Apple Intelligence transcription is the real thing: SpeechAnalyzer and Foundation Models put fast, private, on-device speech-to-text and summarization into macOS 26 and iOS 26, and Notes, Voice Memos, and the Phone app expose slices of it for free. What Apple didn’t ship is the workflow — no meeting capture, no anchored summaries, no export, no library. The system gives you the engine; it doesn’t give you the car.

Dictanta builds the car on Apple’s own engine: meeting recording with no bot, SpeechAnalyzer transcription, Foundation Models summaries anchored to the audio, Markdown export, one library — all on-device, free for your first three recordings, then $9.99/mo, $79.99/yr, or $149.99 lifetime. If you liked the idea of Apple Intelligence transcription but hit its edges in the first week, this is what the same stack looks like finished.