Speech to text

No. The audio is decoded and transcribed in your browser. The only thing downloaded is the speech model itself, once, from Hugging Face; after that it is kept in your browser’s cache.

Base is a good balance (about 75 MB). Tiny (about 40 MB) is the fastest but makes more mistakes. Small (about 250 MB) is noticeably more accurate, especially for Turkish and other non-English languages, but it is slower and needs more memory. On devices with WebGPU the download is larger, but transcription is much faster.

Up to 30 minutes and 500 MB per file. Long recordings are split into pieces of about 30 seconds at quiet moments and transcribed one after another; you can follow the progress and cancel at any time. On slower devices a long recording can take several minutes.

Clear speech with little background noise works best. Names, technical terms and people talking over each other are where mistakes are most likely, so read the text before you use it. Choosing the language instead of “Detect automatically”, and using the Small model, both improve accuracy.

Choose “SRT subtitles” or press “Download .srt”, then load the file in your video editor or player; YouTube, VLC, Premiere, CapCut and most others accept it. To count characters in a subtitle line, use Word & Character Counter.