This page is in your language because of where you are visiting from. View in English
Free & open source · BSD-3-Clause · v2.1.0
Whisper Transcriber Suite runs OpenAI's Whisper model entirely on your own computer — via faster-whisper, whisper.cpp or NVIDIA Parakeet. Drag in audio or video and get a timed transcript back. No cloud, no account, no per-minute cost.
Product tour
Drop a file, pick your options, press Transcribe. Everything else — the queue, downloads, live capture, voice cloning and LAN sharing — sits one tab away. Drag the ring, or use the arrows.
New in 2.1
The app in 42 languages, speaker labels that reach the end of long recordings, files that survive a power cut, and online features that work on every Mac. Below them, what 2.0 added.
2.1 · All platforms
Menus, buttons and messages come in 42 languages (39 on macOS and Linux). The app follows your system language, or you pick one in Advanced settings. Every language except English is a machine translation and is marked as one; Persian, Arabic and Hebrew are offered on Windows for now.
2.1 · All platforms
Speaker names now reach the end of long recordings, speaker detection can no longer crash a transcription, and next to Identify speakers you can set how many people talk (2 to 10) instead of leaving it on Auto.
2.1 · All platforms
Transcripts, exports and subtitles are fully written to disk before they replace the old file, so a power cut cannot leave an empty file. A transcription whose worker stopped after writing all its files counts as finished.
2.1 · macOS
The update check, problem reports and other web lookups no longer fail with certificate verify failed.
2.0 · All platforms
The Transcription Queue tab shows how fast each transcription runs and roughly how long is left, for example about 4.8x, 6 min left. The result card reports the speed, the model and the device when the job is done.
2.0 · All platforms
Tick it on the Download Videos tab and a link becomes <title>-subbed.mp4 with the subtitles burned in: download, transcribe and burn in as one queued job with progress and a working Cancel. The download and the transcript are kept.
2.0 · All platforms
Save a finished transcript as a single .html file that opens in any browser with no install and no internet connection: timestamps, speakers, chapters and a search box. Click a word and the media plays from there.
2.0 · All platforms
One switch under File keeps the app off the network. A line at the bottom of the window says whether a connection is open, and Verify offline now shows test connections being refused. Usage statistics are on by default and can be turned off under Help.
2.0 · All platforms
An Output: English translation option turns non-English speech into English subtitles in one pass, using Whisper's own translate task. It is off by default and not available for every engine or model.
2.0 · macOS
An app menu with Settings (Command-comma), Light and Dark that follow macOS, the system font, files opened from Finder or dropped on the Dock icon, and a notification when a job finishes while the app is in the background.
2.0 · Windows
The taskbar button shows progress, a count of queued and running jobs, and flashes when a job finishes. Title bars follow the Dark theme, and text stays sharp at 125% and 150% display scaling.
2.0 · First launch
A new install asks for your language, Fast or Best quality, and a folder, and shows the model size before anything downloads. An 18-second public-domain clip lets you see a real transcript before you pick a file of your own.
Also new in 2.0: Text to Voice reads texts up to 100,000 characters piece by piece, an Update it button refreshes the video downloader without administrator rights, and a quiet update bar replaces the update dialog. The Video Tiling tab is gone. Everything is in the changelog.
Features
faster-whisper · whisper.cpp · Parakeet
Whisper large-v3 by default, plus large-v3-turbo and distil-large-v3.5. Pick the backend that fits your hardware.
14 formats
Written right next to your input file — plus oTranscribe, ELAN, InqScribe and Express Scribe for professional transcript workflows.
Mic · system audio
Transcribes a microphone — or whatever this machine is playing — as it happens. Cuts are made at natural pauses so words are never split.
Per-word timestamps
Optional "Identify speakers" pass, plus per-word timestamps and time-range clipping for editing precise segments.
Pause · resume · re-run
Live status for every pending and running job, with full control over each row always one click away.
yt-dlp powered
Anything yt-dlp handles, with optional transcribe-on-finish. Downloads resume rather than restart.
Measures before it filters
Measures each recording first and only cleans it when that helps — checks its own output and reverts if it removed speech.
Plus AI tools (summaries, action items, a bilingual SRT pass), an optional offline voice-cloning tab, and a local-network mode — more in the README.
How it works
Transcription runs on your own computer. Your audio is uploaded only if you switch on one of the two optional cloud engines, and transcript text leaves only if you connect a remote AI model yourself.
Audio or video from disk, or anything yt-dlp can fetch. The Tk interface runs in the main process and queues the job instantly.
Every job runs in its own long-lived worker subprocess that keeps Whisper in memory and streams progress back over newline-delimited JSON — so the window never freezes.
A per-worker token and a heartbeat keep the routing robust and let the app detect a stuck worker instead of hanging with it.
Built on OpenAI's Whisper speech-recognition model, run locally through faster-whisper, whisper.cpp or NVIDIA Parakeet, with yt-dlp and FFmpeg for media handling.
Read the full architectureStep-by-step guides: YouTube video to text · video to SRT subtitles · transcripts with speaker labels
Download · v2.1.0 · released 2026-10-12
Everything needed is bundled — a private Python runtime, ffmpeg, ffprobe and yt-dlp. Only the speech model itself downloads later (about 1–3 GB, once); after that, transcription works without an internet connection.
Windows Recommended for you
.exe~238 MB
Most people — normal installer, Start-menu shortcut, upgrades in place. The app isn't code-signed, so Windows SmartScreen may warn once: More info → Run anyway.
Get the installerWindows portable
.zip~360 MB
Unzip and run. No install, no admin rights, happy on a USB stick.
Get the ZIPmacOS Recommended for you
.dmg~275 MB · ~333 MB
Easiest: one line in Terminal — picks the right build and skips the security warning. Or grab the disk image for Apple Silicon or Intel.
One-line installLinux & source Recommended for you
Python 3.11+
git clone https://github.com/Milomilo777/whisper-transcriber-suite.git
cd whisper-transcriber-suite
pip install -r requirements.txt
python gui.py
macOS — one-line install
Downloads the right build for your Mac (Apple Silicon on macOS 14+, otherwise Intel via Rosetta), copies the app into Applications and opens it. Files fetched by curl aren't quarantined, so there's no "unidentified developer" warning to click past.
A=x64; [ "$(uname -m)" = arm64 ] && [ "$(sw_vers -productVersion | cut -d. -f1)" -ge 14 ] && A=arm64; curl -fL -o /tmp/wts.dmg "https://github.com/Milomilo777/whisper-transcriber-suite/releases/download/v2.1.0/WhisperTranscriberSuite-v2.1.0-macOS-$A.dmg" && hdiutil attach -nobrowse -quiet -mountpoint /tmp/wts-dmg /tmp/wts.dmg && cp -R "/tmp/wts-dmg/Whisper Transcriber Suite.app" /Applications/ && hdiutil detach -quiet /tmp/wts-dmg && rm /tmp/wts.dmg && open "/Applications/Whisper Transcriber Suite.app"Prefer the usual way? Open the .dmg and drag the app into Applications; the first launch then needs right-click → Open (macOS 14 and older) or System Settings → Privacy & Security → Open Anyway (macOS 15+).
Every release, with notes, lives on GitHub Releases. Install guide.
Compare
Whisper Transcriber Suite is one of several free, open-source desktop apps that transcribe on your own computer. Here is how it compares with Subtitle Edit, Vibe and Buzz, including when another app fits better. Read the full comparison, with a source for every fact and two more apps, noScribe and aTrain.
| Feature | Whisper Transcriber Suite | Subtitle Edit | Vibe | Buzz |
|---|---|---|---|---|
| Price, licence | Free, BSD-3-Clause | Free, MIT | Free, MIT | Free, MIT |
| Downloads | Windows installer and portable ZIP, macOS (Apple silicon, Intel); Linux from source | Windows (x64, ARM64), macOS, Linux (Flatpak, tar.gz) | Windows, macOS, Linux (.deb, .rpm) | Windows, macOS (1.4.5 is the last for Intel Macs), Linux (Flatpak, Snap, AppImage), PyPI |
| GPU acceleration | NVIDIA (CUDA) | NVIDIA (CUDA), Vulkan | NVIDIA, AMD, Intel (Vulkan / CoreML) | NVIDIA (CUDA), Apple silicon, Vulkan |
| Signed Windows installer | No | Not stated | Yes, per its docs | No |
| Interface languages | 42 (machine-translated) | 35 | 22 | 16 |
| Choose it if | You also want a live microphone, video downloads, 14 output formats and a local-network mode with an OpenAI-compatible route | You want many speech engines (Parakeet, Canary, Qwen3-ASR and more) or a notarized Mac app | You want a Windows installer its docs say is signed, or GPU support that names AMD and Intel | You want Linux packages (Flatpak, Snap, AppImage) or a PyPI install |
FAQ
Something else on your mind? Ask in GitHub Discussions.
Yes. It's BSD-3-Clause licensed, with no subscription and no per-minute cost. The Whisper model itself downloads once (about 1-3 GB) on first launch; after that, transcription works without an internet connection.
No, not by default. Every default backend — faster-whisper, whisper.cpp, NVIDIA Parakeet — runs locally on your machine. Two backends are opt-in only (a cloud Gemini-API backend, and Google Cloud Speech-to-Text), and both stay off until you enable them yourself in Advanced → Backend. Usage statistics (no audio, no file names) are on by default and can be turned off under Help → Usage statistics, the app checks GitHub once a day for a new version, and File → Work offline stops all network use.
Windows, macOS and Linux. Published downloads are a Windows installer, a Windows portable ZIP, and a macOS DMG; Linux runs from source.
srt vtt ass tsv txt json lrc md docx pdf, plus oTranscribe, ELAN, InqScribe and Express Scribe.
Yes — optional speaker diarisation ("Identify speakers"), with per-word timestamps and time-range clipping.
Yes, from the Live tab — a microphone or the system audio, transcribed as it happens.
Yes, any site yt-dlp supports, with an optional transcribe-on-finish step.
Yes, via the optional AI Tools tab in the transcript viewer — summarize, action items, ask-a-question, and a per-segment translate pass that writes a bilingual .srt. It's off by default.
Yes, via the optional Clone Your Voice tab, using OmniVoice (Apache-2.0) entirely on your machine. Off by default; opt in at install time, and you must confirm you have permission to clone the reference voice.
It runs the model locally via faster-whisper by default, so there's no per-minute API cost and no audio leaves your machine — unless you explicitly opt into one of the two cloud backends.
Yes — Web/LAN access mode turns this machine into a transcription page the other devices on the network can use.
Menus, buttons and messages come in 42 languages (machine-translated and marked as such); speaker labels reach the end of long recordings and you can set how many people talk; transcripts, exports and subtitles are fully written to disk before they replace the old file; online features work on every Mac. Version 2.0 added the queue's speed and time left, subtitled video from a link, transcripts as one offline web page, a checkable Work offline mode and translation to English. See What's new in 2.1 and the changelog.
Yes. On the Download Videos tab, tick Make subtitled video: the app downloads the video, transcribes it with Whisper and burns the subtitles in with ffmpeg, producing a file named after the video with -subbed.mp4. The download and the transcript are kept, and an existing -subbed.mp4 is never replaced.
Yes. File → Work offline stops the update check, usage statistics, model downloads and remote AI providers. While it is on, a line at the bottom of the window says whether a connection is open, and Verify offline now shows test connections being refused. With it off, the app contacts the network only for its online settings at startup, the daily update check, usage statistics (which can be turned off under Help → Usage statistics) and downloads you start.
Bugs and feature requests go to GitHub Issues, questions to Discussions, and anything else by email (the address is shown on request in the Contact section). Security reports go through GitHub's private vulnerability reporting.
Contact & support
Press, partnerships, or anything that isn't a bug report.
GitHub Issues
Include your version, OS and the smallest file that reproduces it.
GitHub Discussions
Setup help, ideas, and show-and-tell from other users.
Whisper Transcriber Suite is free and BSD-3-Clause licensed, maintained in the open. If it saved you time, a star on GitHub helps others find it — and a coffee is always appreciated.
Thanks for downloading! While it downloads — if the app ends up saving you time, a ⭐ on GitHub is how other people find it. One click, free.