Free, offline transcription for people with folders full of recordings. Drop in audio files, whole directories, or downloaded YouTube captions; Vertano transcribes and translates everything on your own machine. Nothing is uploaded, nothing is metered, and the transcript lands right next to each source file.

Website · Try it in your browser · Downloads

• Batch transcription. Point it at a year of voice memos. Every audio file (and the audio track of your videos) becomes a .txt transcript saved beside the original. About 100 languages, with first-class Urdu.
• Translation, on device. One toggle turns any spoken language into clean English text. On the Mac app you can also check multiple target languages and get one transcript file per language, translated locally.
• YouTube caption cleanup. Auto-generated captions download with every line doubled. Drop the .srt or .vtt on Vertano and get a clean, correctly timed track, plus translated versions ready to re-upload. Manually authored files pass through untouched.
• Models by capability. Efficient for quick single-language work, Enhanced for accents and noisy rooms, Maximum for multilingual and Indic-language audio with code-switching. You choose; only the small one downloads by default.
• Live recording. The Mac app records from the microphone with a rolling transcript, then saves the audio and text together.
• Private by construction. Transcription and translation run entirely on your machine. Airplane mode works fine.

Betas are unsigned: on macOS, right-click and Open the first time; on Windows, click through SmartScreen.

Known limits, stated plainly: Apple's on-device translation covers about 25 target languages and does not include Urdu, Bengali, Persian, Punjabi, or Pashto. Caption files in those languages still get the cleaned, deduplicated track; audio in them still translates to English through Whisper itself.

This repo holds the Mac app, the Windows/Linux desktop app, the website, and the browser app. Android lives in stenodrop-android; the optional cloud transcription server lives in stenodrop-server.

mac/        Native macOS app (SwiftPM, no .xcodeproj), whisper-cli engine
desktop/    Windows/Linux app (Tauri v2 + Rust, whisper-rs, no ffmpeg)
docs/       Website (GitHub Pages from main:/docs) + browser app (docs/app/)
cli/        Original Python CLI, kept for reference
docs/superpowers/specs/   Design specs, one per feature, reviewed before build

brew install whisper-cpp ffmpeg
cd mac
swift test                # 120+ unit tests
./scripts/e2e-test.sh     # real engine smoke test
./scripts/make-app.sh     # → dist/Vertano.app (ad-hoc signed)

Details, architecture notes, and known limitations: mac/README.md.

cd desktop
npm install
cargo test --manifest-path src-tauri/Cargo.toml
npm run tauri dev

Releases are cut by release.yml on v* tags for all three desktop platforms.

docs/app/ is a static page: Whisper in the browser via Transformers.js plus an opt-in cloud mode against stenodrop-server. Serve docs/ locally with any static file server.

Issues and pull requests are welcome. Feature work here starts from a written spec in docs/superpowers/specs/ and lands with tests; the larger recent features were adversarially design-reviewed before implementation, and the specs record what was decided and why.

MIT licensed. Built on whisper.cpp and OpenAI Whisper. Built by @fahdi.