Use Whisper large-v3-turbo in your browser — no Python, no GPU, no command line. Upload a file, get timestamped text and SRT subtitles.
Transcription runs on OpenAI's open-source Whisper large-v3-turbo, hosted on Cloudflare's GPUs. It's the same model you'd run locally with `whisper --model turbo`, without installing anything or waiting for a GPU queue.
Voice activity detection is on, so long silences are skipped instead of producing the made-up lines Whisper is known for on quiet audio.
The large-v3-turbo model was trained for transcription only — its built-in translate task doesn't work (it returns the original language). So when you tick Translate to English, we transcribe with Whisper first and then translate each timed line with a separate language model. Timestamps stay exactly where Whisper put them.
Running Whisper locally is great if you have a recent GPU and are comfortable with Python and ffmpeg. This page is for everyone else — and for when you just need one file done. Your audio isn't stored, and there's nothing to sign up for on shorter files.
It's OpenAI's open-source Whisper large-v3-turbo model (MIT-licensed), run by us. We're not affiliated with OpenAI; the official source is on GitHub.
No — we only run large-v3-turbo, which is faster than large-v3 with nearly the same accuracy. Smaller models would only be less accurate.
You get phrase-level timestamps (one line per spoken phrase), which is what SRT and VTT subtitles use.
Yes. Without an account you can transcribe files up to 10 minutes, 3 files a day. A free account raises that to 60 minutes a day and 30-minute files; paid plans go up to 60-minute files.
No. Your file is converted in your browser and sent in short pieces for transcription; we don't store the audio or the text. Download your transcript before closing the page.