A browser transcript is useful when you want a searchable draft without sending a recording to a transcription server. In Whisper AI’s free local mode, your device decodes the media and runs the speech recognition model. You still need to review the result, especially names, numbers and quotations.

1. Prepare a short, clear recording

Open Audio to text and choose Upload a file. Select a recording you own or have permission to process. The free local limit is 200 MB and 20 minutes per file. For a longer recording, split it into shorter sections or consider the separate Pro cloud workflow.

Start with a short excerpt if you have not used this device before. MP3 and WAV are useful starting formats; actual decoding support depends on the browser and codec. A file extension alone does not guarantee compatibility.

2. Set the language and model

Under Speech language, select the language you hear, or leave Detect automatically selected. Automatic detection can struggle with very short clips, music or mixed-language speech.

Open Advanced settings to choose a model. Base is the default balanced option. Tiny is a useful first attempt on a slower device. Small needs more memory and a larger download; a larger model does not guarantee an accurate transcript for every recording.

Keep Compatibility (WASM) as the engine unless you know your browser supports WebGPU. GPU acceleration depends on the browser, graphics hardware and available resources.

3. Run Whisper and review the draft

Choose Start transcription. The first run downloads the selected model from Hugging Face. That download is separate from your media: local mode does not upload the recording. Browser caching can make later runs faster, but cached files can be removed when storage is cleared or reclaimed.

Keep the page open while processing. When the draft appears, play the recording and edit each cue’s text. Pay particular attention to proper names, dates, amounts and sections with overlapping voices. Do not treat fluent wording as proof that the model heard it correctly.

4. Export a copy you can keep

Select TXT in Export format, then choose Export for a plain text transcript. Use SRT or VTT when you need timed captions. Local history is stored in this browser; it does not automatically appear on another device. Download an exported copy before clearing browser data.

If something goes wrong

  • The file will not open: try a shorter MP3 or WAV version of the same recording.
  • Processing is slow or stops: close other heavy tabs, select Tiny and use Compatibility mode.
  • The words are in the wrong language: select the spoken language explicitly and try a clear excerpt.
  • A link will not import: use a direct HTTPS media URL that permits browser access, or upload a saved file. A YouTube watch-page URL is not a media file.

For browser storage, cloud history and export details, see the user guide. For the model’s capabilities and limitations, see the official Whisper repository.