Local audio and video transcription for macOS and Windows. Transcriber combines Whisper transcription, optional speaker diarization, a transcript editor, podcast and YouTube import, and Obsidian-ready exports in native desktop apps.
Get the current version from the GitHub Releases page.
| Platform | Recommended download | Notes |
|---|---|---|
| Windows 10 or later | Transcription-Windows-Setup-v*.exe |
Installer with Start menu entry; the easiest option for most people. |
| Windows 10 or later | Transcription-Windows-v*.exe |
Portable app; no installation required. |
| macOS 13 or later on Apple Silicon | Transcription-macOS-v*.zip.part-* |
Join the archive parts before extracting; see below. |
Windows may show a Microsoft SmartScreen warning because the app is not code-signed. Select More info and then Run anyway only after downloading the file from this repository's Releases page.
- Download
Transcription-Windows-Setup-v*.exe. - Run the installer and follow the prompts.
- Open Transcription Windows from the Start menu.
The portable .exe is an alternative when you do not want to install the app: place it in any folder and run it directly.
Download every Transcription-macOS-v*.zip.part-* file and the matching .sha256 file. In Terminal, from the download folder:
cat Transcription-macOS-v*.zip.part-* > Transcription-macOS.zip
shasum -a 256 -c Transcription-macOS-v*.zip.sha256
ditto -x -k Transcription-macOS.zip .Move Transcription macOS.app to /Applications and open it. The app is ad-hoc signed, not notarized; macOS may ask you to confirm the first launch in System Settings.
- Batch queue with file picker and drag-and-drop
- Local audio and video transcription with Whisper Small, Medium, and Turbo
- German, English, and automatic mixed-language recognition
- Optional speaker diarization with automatic or chosen speaker counts
- Speaker names, samples, merging, and post-processing tools
- Transcript editor with search, split/merge/delete, undo/redo, backups, playback loops, and adjustable speed
- Persistent result library with search, archive, trash, and recovered editing state
- RSS, PodcastIndex, and YouTube media import
- VTT import with optional linked media
- Shared glossary context and bilingual terminology review
- Markdown, VTT, SRT, TXT, and CSV export without silent overwrites
- Obsidian-compatible YAML front matter, including source and podcast metadata
The macOS release bundle includes its transcription runtime, models, and media tools, so transcription, diarization, and terminology review work offline after installation.
The Windows release includes the application runtime, FFmpeg/FFprobe/FFplay, Deno, and yt-dlp. Whisper and SpeechBrain model weights are downloaded to %LOCALAPPDATA%\\Transcription Windows\\models the first time they are used, then reused locally. The first model download therefore requires an internet connection and several gigabytes of free disk space.
For video files, Transcriber processes the first audio stream. Files without audio are skipped and reported in the activity log.
The repository keeps platform-specific workflows separate:
Transcription macOS/— shell-based macOS workflows.Transcription macOS App/— native SwiftUI app for Apple Silicon.Transcription Windows/— native Windows desktop app, build scripts, and installer.
cd "Transcription macOS App"
./script/build_and_run.sh --no-runFor a complete distributable bundle, use ./script/build_and_run.sh --bundle --no-run.
cd "Transcription Windows"
install_windows.bat
run_windows_app.batBuild the portable executable and installer with:
build_windows_exe.bat
build_windows_installer.batFrom the repository root:
python3 -m py_compile "Transcription Windows/transcription_windows_app.py" "Transcription Windows/windows_services.py"
python3 -m py_compile "Transcription macOS App/Sources/TranscriptionMacOSApp/Resources/transcribe_bulk.py"
python3 -m unittest tests/test_macos_transcription_backend.py tests/test_windows_services.py tests/test_windows_safe_streams.pyPushing a version tag such as v2.2, or dispatching the Release workflow with a tag, validates shared Python sources and builds both platform packages on GitHub Actions. Each release includes SHA-256 checksums for its published assets.
See NOTICE and THIRD-PARTY-LICENSES.md for attribution and license information.