How to transcribe audio to text
Quick answer: To transcribe audio to text, add your audio, pick a model, choose the language, press Transcribe and fix and download.
- Add your audio. Drop an audio or video file (up to 60 minutes), or record straight from your microphone.
- Pick a model. Fast is fine for clear speech; Better is the sweet spot for most people; Best is for hard recordings if you don't mind a big download.
- Choose the language. Leave it on Detect automatically, or pick it yourself. Tick Translate to English if you want English text from another language.
- Press Transcribe. The first run downloads the model; after that it starts right away. Lines appear as each part is finished.
- Fix and download. Click any line to correct a word, then copy the text or download it as TXT, SRT or VTT.
Is it really private?
Yes, and you can check. The file is decoded by your browser and the Whisper model runs on your own CPU or GPU, in a background worker on this page. The page's security policy only lets it talk to our own site and to Hugging Face, where the model files live, and those requests download the model; nothing goes the other way. Once the model is cached, you could even switch off Wi-Fi and it would still work.
How accurate is it, and how long does it take?
Whisper is very good on clear speech in a quiet room, and gets worse with music, crosstalk, heavy accents or a phone held far away. The bigger models handle those better. Speed depends on your device: with WebGPU on a recent laptop an hour of audio can take a few minutes; on an older phone's CPU expect roughly real time or slower with the Fast model. Long files are cut into 30-second parts at quiet moments, so words don't get chopped in half.
Frequently asked questions
Is my audio uploaded anywhere?
No. The audio is processed on your device and never leaves your browser. The only network traffic is downloading the speech model from Hugging Face the first time; after that it comes from your browser's cache.
How big is the download?
About 40 MB for the Fast model, 77 MB for Better and 250 MB for Best on the CPU. With a GPU the files are larger (about 120–300 MB) because the GPU needs higher-precision weights. You only download each model once.
Which languages does it support?
Whisper understands about 100 languages. The picker lists 36 common ones, including English, Spanish, French, German, Arabic, Urdu, Hindi, Chinese and Japanese, and automatic detection works for those too. It can also translate any of them into English.
Can I make subtitles for a video?
Yes. Drop the video file itself, transcribe it, fix any words, then download SRT or VTT. Most video editors, YouTube and VLC accept both.
Is there a length limit?
Up to 60 minutes per file and 500 MB, because the whole recording has to fit in your browser's memory. For longer recordings, split them into parts with Trim Video first.
Why is it slow on my phone?
Without WebGPU the model runs on the CPU, which is much slower. Pick the Fast model, keep the screen on, and close other tabs. On a laptop with Chrome or Edge it's usually far quicker.
Is it free?
Yes, free with no sign-up, no watermark and no minute limits, because your own device does the work.