# Transcription > How Italic turns what you say into text on your iPhone, in the browser and from your devices, and what is sent where. Every recording becomes text first. Then Italic spells your [Vocabulary](/vocabulary) words your way, [Clean Up](/clean-up) removes filler words if it's on, and a Memo gets a title and a summary. Where each step happens depends on where you recorded. | Step | iPhone, Sudo, Dex and Glyph | Web app and Italic for Mac | | --- | --- | --- | | Audio | Stays on your iPhone | Sent to Italic, then kept only in that browser | | Transcription | On your iPhone | Cloudflare, for Italic | | Clean Up | Apple Intelligence on your iPhone | Italic Cloud | | Title and summary | Your iPhone or Italic Cloud, your choice | Italic Cloud | Italic Cloud is Italic's servers. Wherever you record, only the text syncs to your account. ## On your iPhone [#on-your-iphone] Your iPhone transcribes what you record in the Italic app and what your Sudo, Dex and Glyph record, with speech recognition that runs on the iPhone itself. You'll find it in Profile → Dictation, under **Speech Recognition**. ### Apple Speech [#apple-speech] On iOS 26 or later, Italic uses Apple's on-device speech model. It's on by default: the switch reads **Use Apple Speech**, and **Selected engine** shows what's in use. On an iPhone that doesn't support Apple's advanced speech model, the switch reads **Use Apple Dictation**, and Italic uses Apple's standard dictation model instead. - iOS asks you to allow Speech Recognition for Italic. If it's off, Italic uses a fallback model. To turn it back on, open iPhone Settings → Italic. - Apple may need to download its speech model for your language. You can still record: Italic keeps the audio and transcribes it once the model is ready. ### Fallback speech models [#fallback-speech-models] When Apple Speech can't transcribe a recording, Italic uses a speech model downloaded to your iPhone. That happens on iOS 18 to 25, when Speech Recognition is off for Italic, or for a language Apple Speech doesn't support on your iPhone. With Apple Speech on, find the models under **Fallback speech models**. Without it, they're listed under Speech Recognition, and **Selected engine** reads **FluidAudio**, the engine that runs them. - Each model shows the languages it transcribes, its download size, and whether it gives **Live captions** or transcribes **After recording**. - Tap **Download**, and when it finishes, **Select**. Models stay on your iPhone and can be large; some are over 1 GB. - Italic chooses one that can transcribe your language. If it isn't on your iPhone yet, Italic downloads it the first time a recording needs it. - A model marked **Can't transcribe** your language is never used for it. Your engine and model choices are set on each iPhone. ### Live captions [#live-captions] When the engine supports it, the recorder shows your words as you speak. Apple Speech does. If the live transcript covers the whole recording, it becomes the recording's text. Otherwise, or with a model that transcribes after recording, Italic transcribes the saved audio when you tap **Save as Clip** or **Save as Memo**. Recordings from Sudo, Dex and Glyph are always transcribed from their audio, once your iPhone has copied it from the device. ### Speakers [#speakers] When more than one person speaks in a Memo, your iPhone tells their voices apart and starts each turn on a new line with who's speaking: **Speaker 1**, **Speaker 2** and so on, in the order they first speak. It works this out on your iPhone, with a small model that downloads with the first Memo that needs it, and then transcribes the Memo one turn at a time. - A Memo with one voice has no labels, and neither does a Clip. - A voice heard for only a few seconds, such as someone in the background, and a word or two said over someone else stay in the turn around them. - [Clean Up](/clean-up) leaves the labels as they are. - A Memo saved before your iPhone found speakers gets them when you choose **Transcribe again** on its page, if its audio is still on your iPhone. - If your iPhone can't tell the voices apart, for example because the model couldn't download, the Memo is transcribed as usual, without labels. - Memos recorded in the browser don't have speakers yet. ### When a recording can't be transcribed [#when-a-recording-cant-be-transcribed] Italic doesn't guess. If no speech is found, transcription fails, or no engine on your iPhone can transcribe your language, the recording stays on Home with its audio and a short note, such as "No speech detected". From its **⋯** menu, choose **Retry transcription**, **Save audio** or **Delete recording**. When something needs fixing first, such as allowing Speech Recognition or downloading a model for your language, the recording says so. A change to these settings applies to new recordings. A recording already waiting keeps the settings it was made with. ## Language [#language] The language setting tells Italic what you speak. It belongs to your account, so your iPhone and the web app use the same choice. - **iPhone**: Profile → Dictation → **Language**. - **Web and Mac**: **Dictation** in the sidebar → **Transcription language**. **Automatic** is the default. In the browser, and on an iPhone using a fallback model that knows several languages, Italic detects the language you speak and can switch languages within a recording. With Apple Speech, it transcribes in your iPhone's main language. If transcripts drift into the wrong language, choose yours: English, Chinese, Spanish, Hindi, Arabic, French, German, Japanese, Korean, Russian, Portuguese, Italian, Dutch, Indonesian, Vietnamese, Thai, Turkish, Polish or Ukrainian. Not every engine on your iPhone transcribes every language; the note under Speech Recognition says which one will transcribe yours, and what happens if it can't. ## In the browser [#in-the-browser] The web app records Clips and Memos, in any browser and in Italic for Mac's window. When you choose **Save as Memo**: The audio goes to Italic, and Cloudflare transcribes it for Italic in your transcription language, listening for your Vocabulary words. Italic writes your Vocabulary words your way, as it does on your iPhone. Italic Cloud writes the title and summary and, with **Clean up memos** on, cleans the transcript. The Memo appears on Home. Its audio is kept only in that browser, and only the text syncs. **Save as Clip** stops after the Vocabulary step: no title, summary or clean up, and the audio is deleted once the words are saved. Browser and Mac recordings have no fixed time limit. A long recording is transcribed in smaller pieces and saved as one Memo. Daily transcription allowances still apply; if processing reaches one, the audio stays here to retry later. See [Record](/recording) for saving and retrying. ## Your words, then Clean Up [#your-words-then-clean-up] - **Vocabulary**: the speech engine leans toward your [Vocabulary](/vocabulary) words where it can. Then every transcript, from your iPhone, your devices and the browser, is rewritten to spell them your way. Some fallback models need an extra Vocabulary model of about 100 MB for this; the selected model's row offers **Download**. - **Numbers**: your iPhone writes spoken numbers as digits, so "two hundred" becomes 200. - **Clean Up** runs after transcription. On your iPhone, Apple Intelligence cleans a Clip before it's saved and a Memo just after, and the Memo's summary is written from the cleaned words. In the browser, Italic Cloud cleans the Memo while it writes the title and summary from the words as transcribed. See [Clean Up](/clean-up). ## Titles and summaries [#titles-and-summaries] A Memo gets a title of up to six words and a summary of what you said, in short bullet points. Clips get neither. On your iPhone, choose who writes them in Profile → Personalization → Titles & Summaries → **Process with**. It's set on each iPhone, and covers Memos from your devices too. - **Automatic**, the default: Apple Intelligence on your iPhone writes them when it's ready, and Italic Cloud when it isn't. Italic Cloud also steps in if your iPhone is slow or can't finish, and writes first for a long Memo or one saved while Italic is in the background, with your iPhone as the backup. - **Apple on-device**: only on your iPhone, when Apple Intelligence is ready. Nothing is sent to Italic Cloud. - **Italic**: always on Italic's servers. In the browser, Italic Cloud always writes them. Italic Cloud gets the Memo's text, not its audio, and uses the AI services listed in the [Privacy Policy](https://app.italic.com/privacy). Its summaries group your points under a few headings, and end with **Next steps** only if you said you'd do something. To shape them, fill in **About you** and **Response preferences** in **Personalization** in the web app, or under **About You** in Profile → Personalization on your iPhone. They change summaries Italic Cloud writes, never your transcript. On your iPhone, a new Memo gets its title first, then its summary. If one can't be written, Italic tries again later, and the Memo's page offers **Retry**. To write a title or summary again, see [Home](/home).