Offline AI transcription · v1.0.7
Drop in a recording, meeting, lecture, interview or video and get accurate text, subtitles and a Word document in minutes, in about 100 languages. Andromeda Scribe runs entirely on your own computer, so confidential audio never touches a cloud server.
Built on OpenAI's Whisper speech models, running locally with whisper.cpp, wrapped in an app that is fast to use every day.
Transcription happens on your processor. No uploads, no accounts, no minutes to buy. It even works on a plane.
Automatic language detection, from English, Swedish and Spanish to Japanese and Arabic, plus one-click translation into English.
Record your microphone and, on Windows, the other side of the call too, then transcribe the moment you stop.
Every line is timestamped. Click it to jump the audio there, fix a word in place, and follow along as it plays.
Export clean documents with timestamps to DOCX, TXT, Markdown or JSON, ready for reports, notes and articles.
Create SRT and WebVTT subtitles for YouTube, Premiere, DaVinci Resolve or any video player straight from MP4, MKV or MOV files.
Teach it names, products and jargon so they are spelled right the first time.
No subscription. One license, three computers, a year of updates, and the app keeps working forever.
If you listen back to audio to write things down, Scribe gives you those hours back.
MP3, WAV, M4A, FLAC, OGG, MP4, MOV, MKV, WEBM and more.
Fast drafts with Base, or near-perfect text with Turbo and Large v3.
The transcript appears line by line while it works.
Word, subtitles or text, saved to your Documents folder.
Start free. Upgrade when you want the best accuracy, long recordings and every export format.
Yes. Your recordings are decoded and transcribed on your own computer. The Base speech model is included, so it works offline straight away. The internet is only used to download the larger models (once), to activate a license, and to check for updates.
A 64-bit PC with Windows 10 or 11, or a modern Linux desktop, and a processor from roughly 2015 or newer (AVX2). 8 GB of RAM is plenty for every model; Tiny, Base and Small run fine with 4 GB. On Linux you need WebKitGTK 4.1, which most desktops already have (Ubuntu/Debian: sudo apt install libwebkit2gtk-4.1-0).
On a typical modern laptop, Base transcribes an hour of audio in a few minutes and Turbo (compact) in roughly 10 to 20 minutes. Faster processors with more cores are quicker. You can keep working while it runs.
With the Turbo and Large v3 models, accuracy on clear speech is comparable to the best cloud services. Background noise, crosstalk and heavy accents lower accuracy for every system; the larger models handle them best.
Windows: extract the zip and run Andromeda Scribe.exe. SmartScreen may ask on first launch: choose "More info → Run anyway". Linux: extract the archive and run ./install.sh to add it to your app menu, or start ./andromeda-scribe directly.
If you buy from inside the app it activates itself as soon as the payment is done. Otherwise paste the key from the thank-you page or your e-mail into Settings → License. Lost your key? Recover it here.
Checkout is handled by Stripe: cards, Apple Pay, Google Pay and other local methods. One payment, no subscription. It includes a year of updates, and the app keeps working forever with the version you have.
Not yet. Speaker labels are on the roadmap. Today you get accurate, timestamped text that you can edit right in the app.