Speak to start transcribing
What would you like to do with your transcript?
When enabled, transcripts are automatically saved to your browser's local storage and restored when you reopen the app.
Permanently deletes any transcripts stored in this browser. Does not affect the current session.
When saving or copying the full transcript, each utterance is separated by a blank line.
Only affects file import — live dictation always uses the fastest model. Higher quality uses more memory and takes longer. Changes apply on the next import.
Sorry — Digi Scribe doesn't work reliably on iPhones or iPads yet. It will likely crash after a few sentences.
For the best experience, use a desktop or laptop browser, or an Android phone.
Why? Digi Scribe runs a speech recognition model entirely inside your browser using a technology called WebAssembly (WASM). Each time it transcribes what you say, that model needs to allocate working memory. On most devices this memory is released when it's no longer needed, but on iOS, Apple's browser engine (WKWebView — used by every browser on iPhone and iPad, including Chrome and Firefox) does not shrink WebAssembly memory once it has grown. After a few transcriptions the memory builds up until iOS forcibly kills the browser tab.
This is a platform limitation, not something we can work around without switching to a server-based approach (which would mean your audio leaving your device).
Real-time speech transcription that runs entirely in your browser — no servers, no uploads, complete privacy.
BONUS: Files can also be imported, select quality level in Settings (Low / Medium / High) uses whisper-tiny, whisper-base, or whisper-small
Audio and transcribed text stay on your device. All scripts and libraries are self-hosted — no third-party code is fetched at runtime. The Whisper model is downloaded from HuggingFace on first use only, then cached locally.
If the "Save transcripts to local storage" setting is enabled, transcribed text persists in the browser after the session ends. Use the Clear saved transcripts button in Settings to remove it.
Security audit conducted by AI (Claude, Anthropic, March 2026). Not reviewed by Oxford Information Security.