Transcribe audio or video to text
Turn audio or video into text and SRT or VTT subtitles, in 99 languages, with the Whisper AI running in your browser. Free, no upload.
- Free
- No sign-up
- No file uploaded
- Windows, Mac, Android, iPhone
Loading the tool…
Share this tool
About this tool
An interview to transcribe, a meeting to summarize, a video to subtitle? Drop the file: the Whisper AI listens to the sound and writes the text, with the time of each sentence. You get text to copy and SRT and VTT subtitles, without the sound leaving your device.
When to use it
Transcribe an interview
Get a first draft to correct instead of typing everything while listening.
Subtitle a video
The SRT or VTT file imports into YouTube, editing apps and video players.
Keep a record of a meeting
Transcribe the recording to find a decision or a quote.
How to use
- Choose the mode: fast or accurate.
- Choose the spoken language, or keep automatic detection.
- Drop an audio file or a video.
- Proofread the text, then copy it or download the subtitles.
How it works
Transcription uses Whisper, OpenAI's speech recognition model (MIT license), in its tiny and base versions converted to ONNX. It runs in your browser with ONNX Runtime Web, on several cores when the browser allows.
The sound is split into 30-second windows; each sentence gets its start and end time. The model is downloaded once from zanakit.com, then kept in the browser cache. The sound is never uploaded.
Tips for best results
- Choose the language if you know it: detection can hesitate on a silent start.
- Accurate mode is worth the wait for text you'll publish.
- A recording close to the microphone, without background music, gives far fewer mistakes.
- To keep only one section before transcribing, use “Cut audio”.
Common problems and fixes
The text has mistakes
Switch to accurate mode and choose the language; always proofread names and numbers.
Transcription takes a long time
Processing happens on your device: on a phone, expect roughly the audio's length in fast mode. A computer is much faster.
Sentences repeat
This happens on silence or music: cut those parts before transcribing.
Frequently asked questions
Is my audio uploaded to a server?
No, speech recognition runs entirely in your browser.
How much data is downloaded?
About 54 MB in fast mode or 88 MB in accurate mode the first time; nothing after that.
Which languages are recognized?
Whisper's 99 languages. Quality is better in English, French, Spanish, Portuguese and German than in Indonesian or less common languages.
Are speakers separated?
No, the text doesn't tell who is speaking.
Can I use it commercially?
Yes: Whisper is under the MIT license and the text you get is yours.
Related searches
Still stuck?
Ask on the forum: the category, the tool, your device sheet and the test result will be filled in for you.