Blog · Guides
Voice memos to text in 99+ languages: On-device or Cloud AI?
Most transcription advice is written for English. If you speak Spanish at work and Korean at home, or record a lecture in German and an interview in Arabic, the question is not “can my phone transcribe this” but “which engine should do it, and what happens to the audio”. This guide answers both for Memo, a free app for iPhone on iOS 26 or later.
The two engines, in one paragraph each
On-device AI transcribes with Apple’s speech model and summarizes with Apple Intelligence, on the phone itself. The audio, the transcript and the summary never leave your iPhone, and it works without a connection once Apple’s models are installed. It is available on iPhone 15 Pro and later and every iPhone 16 and 17, with Apple Intelligence turned on, in the languages Apple supports on the device for speech recognition and Apple Intelligence.
Cloud AI uploads the audio to Memo’s servers, where AI providers produce the transcript and summary. It works on every iPhone on iOS 26, in more than 99 languages, with maximum accuracy. Memo asks before every upload, and the Privacy Policy names every provider.
You choose per recording, and Memo tells you what happens to the audio before it happens.
How Memo decides what to offer
When you pick On-device AI, Memo checks three things: the phone (does it have Apple Intelligence), the setting (is Apple Intelligence turned on) and the language (does Apple support it on this device). If any of the three is missing, Memo offers Cloud AI instead. So in a language Apple does not yet support on the device, you are not stuck; you get the cloud engine, which covers more than 99 languages. Which iPhones and languages support On-device AI has the exact conditions.
Which engine for which situation
| You are recording | Pick | Why |
|---|---|---|
| A private conversation in a language Apple supports on your phone | On-device AI | Nothing leaves the phone |
| Anything in a language Apple does not support on the device | Cloud AI | It is the engine that covers that language |
| A long recording on an older iPhone | Cloud AI, with Memo Pro past 30 minutes | Older iPhones do not have Apple Intelligence |
| A recording with no network, for example on a plane | On-device AI | It works offline |
| A recording where accuracy matters more than anything else | Cloud AI | Maximum accuracy, every language |
The app speaks ten languages; the engines speak more
Memo’s own interface is available in English, Arabic, Chinese (Simplified), French, German, Italian, Japanese, Korean, Portuguese (Brazil) and Spanish, and follows your iPhone’s language setting. That is separate from the languages the engines transcribe: Cloud AI handles more than 99, and On-device AI follows Apple’s list. You can run the app in English and transcribe in Turkish, or run it in Japanese and transcribe in French.
Right-to-left languages
Arabic memos are shown right to left throughout the app: the transcript, the summary, the to-dos and the chat.
Mixed-language recordings
People switch languages mid-sentence, and a meeting in Berlin can be two-thirds German and one-third English. Cloud AI handles a switch inside a recording, and the summary usually follows the main language of the recording. On-device AI works best when a recording stays in one supported language. If you know a recording will mix languages heavily, Cloud AI is the safer choice.
Summaries, to-dos and chat in your language
Whichever engine you use, the summary, the title, the to-dos and the chat answers come back in the language of the recording. Ask the chat a question in that language and the answer comes in it, from the words in the recording.
Names, places and specialist words
In every language, the words most likely to come out wrong are proper names, place names and technical vocabulary, because the models have heard them least. The transcript is timestamped for exactly this: tap the line, hear the original, fix the name. For anything you will quote, check.
Getting the text out
From the memo menu you can copy the transcript or the summary, share the text, print, export a PDF, keep the audio, or share the memo as a private link that opens in a browser in any language. See Export a memo.
The short version
Speak whatever language you speak. On-device AI when your phone and the language allow it and privacy matters most; Cloud AI when you need any language, any iPhone, or the best accuracy. Memo checks the conditions for you and tells you what happens to the audio every time.
Try it on your own recordings.
Memo is free on the App Store. Requires iOS 26 or later.
