Any media into any.
Speech into text.All on your computer.
MediaChef is a free desktop app that turns FFmpeg and Whisper into recipe cards: convert video and audio, pull the sound out of a clip, transcribe a recording to text — offline, on your own machine, with nothing uploaded anywhere.
Free and open source — read the code on GitHub
- Version 0.4.0
- 17 built-in recipes
- 4 Whisper models
- macOS · Windows · Linux
- Open source · GPL-3.0
How it works
-
Drop a file
Drag in a video or a recording. MediaChef probes the file and shows only the recipes that fit what it found inside.
-
Pick a recipe
Every action is a plain card: “Extract audio to MP3”, “Make SRT subtitles for a video”, “Compress video”. Defaults are already sensible; the live preview shows the exact FFmpeg command.
-
Take the result
The file lands next to the original — or in a folder you choose. The queue shows progress, time left and where the output went.
Recipes instead of commands
FFmpeg can do almost anything — in the language of the terminal. MediaChef translates: you choose what to do, the parameters are already set. A live preview shows the real command, so you learn as you go.
Extract audio to MP3
Pull the soundtrack out of any video.
Make SRT subtitles for a video
Whisper listens and writes an SRT file.
Compress video
Fit a clip into a messenger-friendly size.
Video to GIF
A sharp looping GIF, 10 to 24 fps.
Transcribe audio to text
Meeting or voice memo into plain text.
Translate speech to English
Whisper transcribes and translates in one pass.
Convert MP4 to MKV
Repackage without re-encoding — instant.
Remove audio from video
Drop every audio track, keep the picture.
What it does out of the box
Seventeen recipes ship with the app. Each one is a real FFmpeg or Whisper job with the parameters already filled in, and each one writes its result next to your file as {name}.{what}.{ext}.
| Direction | Recipe in the app | What you get |
|---|---|---|
| Video → audio | «Extract audio to MP3» | clip.audio.mp3 at 128, 192 or 320 kbps |
| Video → subtitles | «Make SRT subtitles for a video» | clip.subs.srt with timings |
| Audio → text | «Transcribe audio to text» | talk.transcript.txt, plain text |
| Any language → English | «Translate speech to English text» | talk.english.txt in one pass |
| MP4 → MKV | «Convert MP4 to MKV» | Repackaged, not re-encoded — seconds, no quality loss |
| Video → smaller | «Compress video (quality preset)» | H.264 at CRF 23, 28 or 33 |
| Video → GIF | «Video to GIF» | 10–24 fps, width 320–640 |
| Anything else | «Custom FFmpeg command» | Your own arguments, with the command preview |
Transcribe audio to text without leaving your device
Whisper, running locally. OpenAI's speech model executes on your own processor via whisper.cpp — recordings never leave the machine.
Models downloaded in-app. From the 78 MB tiny to the 1.62 GB large-v3-turbo — pick per task on the Models screen.
Text or subtitles. Plain TXT, SRT and VTT with timestamps, or JSON with segment times for tooling.
Honest results. If a file has no speech, MediaChef says “No speech detected” — it never ships an empty file with a green checkmark.
Whisper models
Transcription runs on whisper.cpp with OpenAI's Whisper models. You download a model once from the Models screen; after that everything works with the network off. Bigger model, better text, slower run.
| Model | Download | What it's for |
|---|---|---|
tiny | 78 MB | Fastest, rough quality. A draft of clear speech in a fraction of the time. |
base | 148 MB | Fast, ok quality. Good enough to search a recording for the bit you need. |
small default | 488 MB | Recommended balance — the default in every transcription recipe. |
large-v3-turbo | 1.62 GB | Best quality, tuned for Apple Silicon. For text you intend to publish. |
Your files never travel
An online converter asks you to upload the file to someone else's server, wait in a queue and trust their retention policy. MediaChef works on your CPU: a gigabyte of screen recording and a private meeting audio convert the same way — with Wi-Fi switched off.
- No uploads
- No size limits
- No subscriptions
Open source, engines included
GPL-3.0, with the whole development history public on GitHub. Since 0.4.0 every download carries its own engines — nothing to install in PATH, nothing to configure.
- FFmpeg 9.0.1
- conversion · GPL v3
- whisper.cpp v1.7.6
- speech recognition · MIT
- macOS · Windows · Linux
- dmg ~66 MB · installer ~82 MB · AppImage ~181 MB · deb ~118 MB
Exact versions and licenses of everything bundled — NOTICE.md
Questions, answered
Is it really free?
Yes. MediaChef is open source under GPL-3.0 — no account, no trial, no watermark. The converter and the transcription are the whole app, and the code is public on GitHub.
Which formats are supported?
Everything FFmpeg reads: MP4, MKV, MOV, WebM, AVI, TS, MP3, WAV, FLAC, M4A, OGG and dozens more. MediaChef probes the file with ffprobe instead of trusting its extension, so the recipes you see are the ones that actually fit it.
Where do my files get uploaded?
Nowhere. Conversion and transcription run entirely on your computer. The only thing MediaChef ever downloads is a Whisper model — once, on the Models screen.
How accurate is the transcription?
It uses OpenAI's Whisper models through whisper.cpp v1.7.6. Accuracy scales with the model you pick: tiny is instant and rough, small is the balanced default, large-v3-turbo is near-human on clear speech. The language is detected automatically.
Does it work offline?
Yes. FFmpeg and Whisper ship inside the download, so conversion works offline from the first launch, and transcription works offline after you fetch a model once.
Why not just use an online converter?
For one small public file, an online converter is fine. Private recordings, multi-gigabyte videos, batches and anything under an NDA are better done locally: no upload wait, no size cap, and nothing sitting on someone else's disk.
Put a chef in charge of your media
Free, open source, 3 platforms. Version 0.4.0.
MediaChef is young: builds are not yet signed by Apple or Microsoft, so the first launch asks for confirmation — a plain-text how-to ships inside every download.