On-device transcription on iPhone: which models, which languages, and why it matters

For most of the last decade, transcribing a recording meant sending it to a server. That was the only place with enough computing power to do the job well. Newer iPhones changed that. With Apple Intelligence, an iPhone can turn speech into text, and text into a summary, without the audio leaving the phone.

This post explains what that means in plain words, which iPhones can do it, which languages it covers, how Memo uses it, and when Cloud AI is still the better choice.

What “on-device” means

On-device processing means the work happens on the chip inside your iPhone. A speech model stored on the phone listens to the audio file and writes the text. Apple Intelligence, also on the phone, reads the text and writes the summary. At no point is the recording uploaded. No server receives it, no account is needed to identify you, and it makes no difference whether you have a signal.

Cloud processing is the reverse. The audio is uploaded to a server, the server does the work, and the text comes back. It is how most transcription has worked for years, and it is why cloud services can cover so many languages. But it means a copy of your recording exists somewhere other than your phone, at least while it is being processed.

The difference matters most for recordings you would not want to email to a stranger: a medical appointment, a conversation with a lawyer, a meeting about someone’s pay, a private note to yourself. For those, on-device is the option where the question “where did my audio go?” has a one-word answer: nowhere.

Which iPhones support it

On-device AI in Memo needs Apple Intelligence, and Apple Intelligence runs on these phones:

iPhone 15 Pro and later
every iPhone 16 model
every iPhone 17 model

Phones before that, including the standard iPhone 15 and older models, do not have Apple Intelligence, so On-device AI is not available on them. Memo still works on every iPhone running iOS 26, using Cloud AI.

Having the right phone is not quite enough. Apple Intelligence also has to be turned on. Open Settings, go to Apple Intelligence & Siri, and turn Apple Intelligence on. Once it is on, Memo can use it.

Languages follow Apple’s list

Apple Intelligence, and the speech model that comes with it, work in the languages Apple supports on the device. Memo does not control that list. Apple adds languages over time, and when a language arrives on your phone, On-device AI in Memo can use it.

If your language is not on Apple’s list yet, you are not stuck. Memo offers Cloud AI in other languages, and Cloud AI handles more than 99 of them. The same is true if you record in several languages and some of them are not supported on the device.

Memo’s own interface is separate from this and is available in English, Arabic, German, Spanish, French, Japanese and Brazilian Portuguese.

How Memo uses it

When you choose On-device AI in Memo, three things happen on your phone and nowhere else.

The transcript. Apple’s speech model turns the recording into text.
The summary. Apple Intelligence reads the transcript and writes the summary.
The chat. When you ask a question about the memo, the answer is worked out on the phone as well.

You still get a title and an emoji for each memo, so a memo made with On-device AI looks and works the same as one made with Cloud AI. The recording is stored on your phone, the text is stored on your phone, and you can copy, share, print, or save the text as a PDF from the memo menu whenever you want.

Two more things follow from all of this. You do not need an account for On-device AI, because there is no server to sign in to. And it works without a connection, so it does exactly the same thing on a plane as it does at your desk.

The honest trade-offs

On-device AI is the right default for a lot of people, but it is not the right choice for everyone, and it is worth being clear about why.

Languages. Cloud AI covers more than 99 languages. On-device AI covers the languages Apple supports on the device, which is a shorter list. If your language is not there, Cloud AI is the answer.

Longer recordings. On the phone, your own chip does the work, so a long recording takes longer to process than a short one and uses battery while it runs. Cloud AI does that work on a server instead. On the other hand, On-device AI has no upload limits, because nothing is uploaded, while on the free tier Cloud AI recordings can be up to 30 minutes long and there is a daily allowance. A Memo Pro subscription lifts those limits.

Older phones. If your iPhone does not have Apple Intelligence, On-device AI is simply not available. Cloud AI works on every iPhone running iOS 26.

When you use Cloud AI, here is what happens to the audio. It is uploaded to Memo’s servers and processed by AI providers. You need to sign in with Apple or Google, and the app asks before every upload, so nothing is sent by accident. Memo never uses your audio or your text to train AI models. The privacy policy has the full details.

How to check which engine a memo used

The engine is a choice you make in Memo, and it applies to the memos you process. There are two quick ways to be sure what happened to a particular recording.

First, the upload prompt. Cloud AI asks before every upload. If you were never asked to upload a recording, it stayed on your phone.

Second, your account. Cloud AI needs Sign in with Apple or Google. If you have never signed in to Memo, nothing has been uploaded, because there is no way for Cloud AI to run without an account.

If you are still unsure, the help pages cover Memo’s settings, and the support page explains how to get in touch.

Frequently asked questions

Does On-device AI need an internet connection?

No. Transcription, summary and chat all run on the phone. You can record and process memos with no signal at all.

Do I need an account for On-device AI?

No. On-device AI works without an account, because there is no server involved. Cloud AI needs Sign in with Apple or Google, since the audio is processed on Memo’s servers.

Is my audio used to train AI models?

No. Memo never uses your audio or your text to train AI models, whichever engine you choose. With On-device AI the audio never leaves your phone at all.

On-device AI gives you a transcript, a summary and a chat with the audio staying on your phone, and Cloud AI is there for every other language and every other iPhone. If you have an iPhone on iOS 26 or later, get Memo on the App Store and choose the engine that fits the recording.