Turn speech in a video into an editable transcript and subtitles, transcribed on your own device. A strong first draft, with its limits stated plainly.
Your files are processed in your browser, on your own device. Nothing is uploaded to a server.
Turn the spoken words in a video into text you can read, search, quote and edit: a transcript as a plain text file, or as subtitles in SRT or VTT format with timings. The speech recognition runs on your own device, in your browser, which means a private recording can be transcribed without its audio ever reaching a server.
Honesty first, because this category needs it: browser transcription is draft quality. Server services with data-centre hardware advertise up to 95 percent accuracy. A model small enough to run in a tab will do well on clear, close speech and struggle with overlapping voices, heavy accents, background music and technical vocabulary. Read this transcript as a strong first draft, fine for finding moments, drafting notes and starting subtitles, not as a verbatim legal record.
1. Drop the video into the box. The sound is extracted and prepared as the 16 kHz mono audio speech models expect.
2. Choose the model. Tiny, about 43 MB, is the fast default and suits clear speech. Base, about 80 MB, is slower and noticeably better on ordinary recordings. The size is shown before anything downloads.
3. Pick the language, or let the model detect it. Press Transcribe and keep the tab in front of you.
4. Read and fix the transcript in the editor, then download it as TXT, or as SRT or VTT subtitles with timings.
Distance and noise, in that order. A phone on the speaker’s desk transcribes beautifully. The same phone across a hall gives the model echo and murmur to guess through. One speaker at a time beats a panel. Naming names and jargon are the known weak spots, so check proper nouns while you edit. The model files download once per model and are cached, like the video engine.
Thirty minutes of video is the recommended ceiling and sixty is the hard cap, because transcription here runs many times slower than the recording on ordinary hardware. There are no speaker labels: the model hears words, not who said them. There is no punctuation guarantee beyond what the model produces, and the editor is there because a pass with your eyes is part of the process, not a failure of it. The first stage uses the same in-browser video engine as the rest of the family: only the words are extracted, fully on your device, like everything in the video tools collection. If you only need the sound itself, the video to audio converter stops one step earlier.
Server transcription is usually better, and this page says so above. What it cannot offer is this: an interview recorded in confidence, a medical consultation, a legal deposition, a child’s school recording, transcribed without the audio leaving the room. For recordings where the content is the sensitive part, a draft made on your own hardware beats a polished one made on a stranger’s server, and that trade is the reason this tool exists in a browser at all.
On clear, close, single-speaker audio, good enough to read and work from with light fixes. On noisy, distant or overlapping speech, treat it as a rough draft. It will not match a server service running large models, and this page will not pretend otherwise.
Tiny for a quick draft of clear speech and for slower devices. Base when the recording matters and you can wait longer. It is roughly double the download and clearly better on ordinary audio.
No. The output is one stream of words with timings. If you need who said what, add the names during your editing pass.
SRT and VTT, the two formats players, editors and platforms accept, plus plain TXT for the words alone. The transcript can be edited before downloading.
Up to 60 minutes, with 30 recommended. Transcription in a browser runs much slower than real time on typical hardware, so a half hour recording is a long job. Keep the tab open and visible.
No. The model downloads to your browser and listens there. The audio and the transcript never leave your device, which makes this the right choice for private recordings.
Same rules as this one: free, in your browser, with nothing uploaded.
Take the sound out of any video as an audio file.
The classic job. Video in, MP3 out.
Move between MP4, WebM and MOV without a fight.
A smaller video for WhatsApp, email and uploads.
Cut the dead seconds from the start and end.
Turn a short clip into a GIF that loops.
Any video file in, a plain MP3 of its sound out.
Join clips end to end into one video file.