How to transcribe audio or video to text — free, with no upload
Transcription sites charge by the minute and want your recording on their servers — awkward when it's a client call, a doctor's dictation or an interview under embargo. The same AI they use (OpenAI's Whisper) runs happily inside a modern browser. Here's how to use it without the invoice or the upload.
Step by step
- Drop your file into the tool. Any common audio or video works — MP3, WAV, M4A, MP4, MOV, WEBM. The audio track is extracted on your device by FFmpeg compiled to WebAssembly; a video never needs converting first.
- Pick a model and language. The fast model is built into the site and even works offline. The higher-accuracy model downloads once (~74 MB, then cached) and is noticeably better with accents, noise and technical vocabulary. Language is auto-detected across 90+ languages; setting it manually helps on clips under a minute.
- Transcribe and export. Progress is shown as each 30-second window is decoded. Read the transcript on the page, then download it as plain text (TXT), SRT subtitles or WebVTT captions — timestamps included.
Worth knowing
Speed scales with your hardware. A browser with WebGPU (recent Chrome or Edge on a machine with any GPU) is several times faster than plain CPU mode — worth trying for long recordings. And because the transcript never existed anywhere but your device, there's nothing to delete from anyone's servers afterwards.
Frequently asked questions
Is there really no minutes limit?
Really. Transcription services charge per minute because their servers do the work. Here your device does the work, so an hour-long recording costs nothing — it just takes longer.
Is the recording uploaded anywhere?
No. The Whisper model runs inside your browser, and the site's Content-Security-Policy pins down exactly which hosts a page can talk to. Confidential meetings and interviews stay on your machine.
How do I get subtitles for YouTube?
Download the SRT file and upload it in YouTube Studio under Subtitles — or the VTT file for players that use WebVTT. Timestamps are included automatically.
How accurate is it?
Clear speech in a major language transcribes very well even on the fast model. Heavy accents, crosstalk and far-away microphones are where the higher-accuracy model earns its download.