Unlimited transcription in your browser with Whisper AI
Choose an audio file and the Whisper model runs inside your browser to transcribe it. There is no need to set up Python, ffmpeg or GPU drivers. The audio never reaches a server, and there is no account or payment.
You can normally leave this on Auto. Change it only if the result clearly has the wrong number of speakers
No file? Record with your microphone. Transcription starts after you stop recording (up to about 000 min on this device with the selected model). Your recording is not sent to any server either.
How to use it
- Choose a model. The default is picked automatically from what your device can handle. On a PC with WebGPU, Large v3 Turbo offers a good balance of accuracy and speed.
- Choose the language. Auto-detect works, but specifying the actual language reduces recognition errors.
- Choose an audio or video file. Transcription starts as soon as the file is loaded, and when it finishes you can choose TXT, Word, SRT or another format under "Download".
About Whisper running in the browser
Whisper is a speech recognition model released by OpenAI. It is normally used by setting up a Python environment and installing ffmpeg and GPU drivers. This tool runs the same Whisper models converted to a format that works in the browser. A model is downloaded the first time, and after that the copy stored in your browser is used.
Six models are available: Tiny, Base, Small, Medium, Large v3 Turbo and Large v3. On a PC with WebGPU they run on the GPU; on phones and devices without WebGPU they run on the CPU. Larger models are more accurate but need a bigger download and more memory. Large v3 Turbo keeps the Large v3 encoder and uses a smaller decoder, so it runs fast with accuracy close to Large v3. For the error rates and processing times we measured for each model in each language, including models other than Whisper, see Model comparison and how to choose.
Whisper sometimes writes out sentences nobody said during silence or music. Repeated sentences are removed automatically, but a sentence that appears only once remains, so delete that line on the results screen.
If something goes wrong
- If processing stops partway, your device may be running out of memory. Switch to the next smaller model or split the audio into shorter parts.
- If proper nouns or technical terms are often missed, switch to Small or a larger model.
Frequently asked questions
Do I need Python or a GPU?
No. A recent browser such as Chrome or Edge is enough. On a PC with WebGPU it runs fast on the GPU, and on phones or devices without a usable GPU it runs on the CPU (it takes longer).
Which model should I choose?
The default is chosen automatically from what your device can handle. On a PC with WebGPU, Large v3 Turbo offers a good balance of accuracy and speed; on a phone, choose Tiny or Base.
Are the results the same as the Python version of Whisper?
The model is the same Whisper, but it is made lighter to run in the browser, so fine details of the output may not match the Python version.
Does it work offline?
Yes. Once the model has been downloaded, you can transcribe without a network connection.
Getting the audio ready, and after transcribing
Help us improve the service
Sending us the circumstances of a run — successful or failed — helps us track down problems (a failure rate needs both). Knowing which devices and settings fail most often lets us revise the recommended defaults and fix the bugs behind them. If you don't mind, leaving this on is a real help.
What is sent
Your device specs (GPU model, memory, core count), browser and OS, the model and language you chose, your speaker diarization setting (on or off, and the number of speakers you set), the page's display language, how far the run got (including when you stopped it or closed the page), the error type and details, the file's format with a rough size and length, and how long it took. A temporary number that links runs in the same tab (it is deleted when you close the tab). On the result screen: how many times you used editing (such as renaming speakers or splitting lines), translation, summaries and subtitle imports, the formats you chose under “Download”, how often you used “Copy text” and “Download all”, and the translation language pair.
What is never sent
Audio, transcripts (including speaker names and translations), file names, and IP addresses. The data goes only to this site's own server, never to a third-party service.