Voice to Text Offline

Transcribe turns speech into text, and text into another language — entirely on your iPhone or iPad. Record a meeting, a lecture or a voice memo and get a transcript. Import an audio file instead. Translate the transcript, read the translation beside the original, have a summary written for you, or ask the recording questions. NOTHING LEAVES YOUR DEVICE • No account, no sign-in, no subscription • No servers of ours: your audio and your text are never uploaded • Transcription needs no network at all, so it works offline and in airplane mode • The app collects no data — that is the whole privacy policy TRANSCRIPTION, ON DEVICE • Runs on the Neural Engine with Parakeet, built into the app — 25 European languages, no download • Need more languages? Download the optional Whisper large-v3-turbo model once for 99 languages • Live preview while you record, then a full-quality pass over the finished audio • Also transcribes audio files you import from Files or share from another app • Fix anything it mishears: the transcript is yours to edit TRANSLATION • Translate a transcript into the language you choose • Translation runs on device, and by default uses Apple's built-in translation — so there is nothing to download before your first translation • If the system is missing a language pack, the app says which one and where to get it • Prefer no limits on language? Download the optional on-device model once and every language the recogniser can hear becomes a translation target, direct between any two SUMMARY AND QUESTIONS • Ask the on-device model for a summary of a long recording: a few sentences, saved with the transcript • Ask questions about a recording and get answers drawn from what was actually said, in a conversation you can keep • Both are written in any language you pick • Both use the optional on-device model, which is installed once from Settings • Bring your own model instead: import any .gguf file you already have, and translation, summaries and questions all use it WHAT TO EXPECT • Apple's built-in translation covers the languages the system supports; the optional model covers all 99 the recogniser can hear • The optional on-device model is about 1.22 GB, downloaded once over Wi-Fi, and after that works offline permanently • The app is about 610 MB, and about 1.8 GB with the optional model installed • Automatic transcription, translation, summaries and answers are good but not perfect. They can mishear names, numbers and technical terms, particularly with background noise, strong accents or several people talking at once. Please check anything that matters • Transcribe is not a substitute for a professional interpreter or translator REQUIREMENTS • iOS 18.0 or later ATTRIBUTION Speech recognition uses Parakeet (CC BY 4.0 licence, © NVIDIA and moondream) or OpenAI's Whisper (MIT licence, © OpenAI). Summaries and questions use Qwen3 (Apache License 2.0, © Alibaba Qwen team). Everything runs locally on your device; no model is modified.