faster-whisper · whisper.cpp · Parakeet
Local, private transcription
Whisper large-v3 by default, plus large-v3-turbo and distil-large-v3.5. Pick the backend that fits your hardware.
Free & open source BSD-3-Clause
Sound in. Text out. Nothing leaves your machine.
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.
Windows macOS Linux
- 14transcript formats
- 3local speech engines
- 0uploads by default
- $0no subscription, no per-minute cost
Everything you'd want from a transcription tool. Running on your machine.
14 formats
Every transcript format you'll need
Written right next to your input file — plus oTranscribe, ELAN, InqScribe and Express Scribe for professional transcript workflows.
Mic · system audio
Live transcription
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
Speaker diarisation
Optional "Identify speakers" pass, plus per-word timestamps and time-range clipping for editing precise segments.
Pause · resume · re-run
Batch queue
Live status for every pending and running job, with full control over each row always one click away.
yt-dlp powered
Video & audio downloads
Anything yt-dlp handles, with optional transcribe-on-finish. Downloads resume rather than restart.
Measures before it filters
Adaptive denoise
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, a local-network mode and video tiling — full list in the README.
One window. Every step.
Drop a file, pick your options, press Transcribe. That's the whole workflow — four more tabs cover queueing, downloads, video tiling and LAN sharing.
A drop target, engine and language pickers, speaker-label and word-timestamp options.
Private by architecture. Not by promise.
Every default engine runs in processes on your own computer. Nothing is uploaded unless you switch on one of the two opt-in cloud backends yourself.
-
01
Drop a file, or paste a link
Audio or video from disk, or anything
yt-dlpcan fetch. The Tk interface runs in the main process and queues the job instantly. -
02
A worker holds the model
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.
-
03
The transcript lands next to your file
A per-worker token and a heartbeat keep the routing robust and let the app detect a stuck worker instead of hanging with it.
Download. Transcribe tonight.
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 the app is fully offline.
Windows Recommended for you
Setup-Standard.exe
~215 MB
Most people — normal installer, Start-menu shortcut, upgrades in place.
Get the installerWindows portable
Portable.zip
~330 MB
Unzip and run. No install, no admin rights, happy on a USB stick.
Get the ZIPmacOS Recommended for you
macOS-*.dmg
~400 MB
A disk image for macOS — pick the build on the release page that matches your Mac.
Get the DMGLinux & source Recommended for you
Build from source
Python 3.11+
git clone https://github.com/Milomilo777/whisper-transcriber-suite.git
cd whisper-transcriber-suite
pip install -r requirements.txt
python gui.pyEvery release, with notes, lives on GitHub Releases. Install guide.
Frequently asked questions
Something else on your mind? Ask in GitHub Discussions.
Is Whisper Transcriber Suite free?
Yes. It's BSD-3-Clause licensed, with no subscription, no per-minute cost and no telemetry by default. The Whisper model itself downloads once (about 1-3 GB) on first launch; after that the app runs fully offline.
Does it upload my audio or video anywhere?
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.
What platforms does it run on?
Windows, macOS and Linux. Published downloads are a Windows installer, a Windows portable ZIP, and a macOS DMG; Linux runs from source.
What transcript formats can it produce?
srt vtt ass tsv txt json lrc md docx pdf, plus oTranscribe, ELAN, InqScribe and Express Scribe.
Can it label who is speaking?
Yes — optional speaker diarisation ("Identify speakers"), with per-word timestamps and time-range clipping.
Can it transcribe a live microphone or system audio?
Yes, from the Live tab — a microphone or the system audio, transcribed as it happens.
Can it download and transcribe online videos, e.g. from YouTube?
Yes, any site yt-dlp supports, with an optional transcribe-on-finish step.
Can it summarize a transcript, pull out action items, or answer questions about it?
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.
Can it clone a voice and make it say new text?
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.
How is this different from calling the Whisper API directly?
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.
Can several people on the same network use it without installing it on every machine?
Yes — Web/LAN access mode turns this machine into a transcription page the other devices on the network can use.
Support the project
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.