Free, unlimited, on-device
Edit video by transcript
Drop in a video or podcast, get a word-timed transcript, then delete words to cut the media. Transcription, filler-word removal and export all run inside this browser tab, so nothing is uploaded and there is nothing to sign up for.
First transcription downloads the speech model (about 200 MB for Whisper Base, about 600 MB for Small) and caches it. A Chromium browser on a desktop is recommended.
Long file, tricky audio, or no patience for a model download?
This editor runs a small Whisper model on your own machine, which is the price of it being free and private. When you need a two-hour recording, a language other than English, clean speaker separation or accuracy you can publish, transcribe it on ConvertAudioToText instead, export SRT or VTT, and import that transcript here to do the cutting.
How it works
Four steps, all of them local. There is no queue, no upload progress bar and no job to wait on.
1. Drop in your file
Video or audio: MP4, WebM, MOV, MKV, MP3, WAV, M4A. The file is read straight off your disk by the browser. It is never uploaded.
2. It transcribes on your machine
Whisper runs in a Web Worker on your own CPU or GPU and returns a transcript with a timestamp on every single word, grouped by speaker.
3. Delete words, the video cuts
Select text, press backspace, and the matching stretch of media is cut. Playback skips the cuts live so you hear the edit before you export.
4. Export the final cut
ffmpeg.wasm trims and re-joins the kept ranges in the browser and hands you an MP4 (or M4A for audio) to download.
What people use it for
Remove filler words from a video online
One click strips every detected um, uh, like, you know and so on. Review the flagged words first if you want, then export.
Cut video by deleting text
No timeline scrubbing. Read the transcript, delete the sentence you do not want, and the cut lands frame-accurately on the word boundary.
Free podcast editor in the browser
Audio files work exactly like video. Trim tangents, kill dead air, export an M4A. Speaker labels make multi-guest episodes readable.
Nothing is uploaded
There is no server-side pipeline behind this page. Your file is opened by the browser, decoded in WebAssembly, transcribed by a model running on your own hardware, and exported back to a file on your disk. We could not read your media if we wanted to. Recent projects are kept in your browser's IndexedDB and you can delete them from the list at any time.
We do record anonymous product analytics on our own self-hosted PostHog: page views and four events telling us the tool loaded, a transcription started, an export finished, or a link out was clicked. No file names, no audio, no transcript text.
What it needs from your machine
- A Chromium desktop browser is recommended. The page is served cross-origin isolated so SharedArrayBuffer is available for multi-threaded WebAssembly.
- WebGPU is used when your machine exposes it; otherwise it falls back to WebAssembly, which works but is slower.
- Everything is held in RAM, so very long files can exhaust the tab. Short clips are where this tool is happiest.
Frequently asked questions
Is CATT Editor really free?
Yes. There is no account, no sign-up, no export limit, no watermark and no paid tier of this editor. All the work happens on your own machine, so there is no per-minute cost for us to pass on to you. We pay for the page, you pay for nothing.
Does my video get uploaded anywhere?
No. Your browser reads the file locally, decodes it with ffmpeg.wasm, and runs the speech model on your own hardware. The media, the audio and the transcript never leave the device. The only network requests the page makes are the app itself, the one-time AI model download from Hugging Face, and anonymous product analytics that contain no file data.
Why does the first transcription download several hundred megabytes?
Because the speech model runs on your machine, it has to get there first. Whisper Base is roughly 200 MB and Whisper Small is roughly 600 MB, pulled from the Hugging Face CDN the first time you transcribe and then cached by the browser. After that first run, transcription and export work with the network disconnected. If you would rather not download a model at all, transcribe the file on our servers at ConvertAudioToText and import the transcript here.
Which browser should I use?
A recent Chromium browser (Chrome, Edge, Brave, Arc) is the recommendation. The editor needs SharedArrayBuffer, which requires the cross-origin isolation headers this site sends, and it uses WebGPU when your machine exposes it, falling back to WebAssembly otherwise. Firefox works but is usually slower. Safari and most mobile browsers are not reliable for this workload.
How long a file can I edit?
We do not impose a limit, but your browser does. Everything is held in memory, so a laptop is comfortable with clips up to roughly ten to twenty minutes and gets slow or runs out of memory beyond that. For a two-hour interview, transcribe it at ConvertAudioToText, download the SRT, and import it here to do the cutting.
Can I bring my own transcript?
Yes. Pick Import transcript on the source menu and load an SRT, VTT or JSON caption file. Skipping Whisper means no model download and no waiting, and it is the fastest path for long recordings.
How accurate is the transcript?
Whisper Base and Small are the two models small enough to run in a browser, so they trade accuracy for size. They are good on clean speech and get noticeably worse with accents, crosstalk, background noise or non-English audio. For a transcript you intend to publish, use a larger model on ConvertAudioToText and import the result.
What formats can I export?
Video exports as MP4 (H.264 video, AAC audio). Audio-only projects export as M4A. Cuts are re-encoded rather than copied, so they land exactly on the word boundary instead of the nearest keyframe.