Short answer: WhisperWeb is independent

WhisperWeb (whisperweb.site) is an independently developed website and product. It runs OpenAI’s open-source Whisper model locally in your browser, but it is not made, operated, endorsed or sponsored by OpenAI, and it is not an official OpenAI product. Whisper is the name of the open-source speech recognition model that OpenAI released; the WhisperWeb name describes what this site does: it runs Whisper on the web. It is also not connected to other websites or apps that use similar names.

What WhisperWeb is built on

Speech recognition uses OpenAI’s open-source Whisper model in its tiny, base (default) or small size. The site loads ONNX versions of these models published by onnx-community on Hugging Face and runs them with Transformers.js and ONNX Runtime Web, using WebGPU when your browser supports it and WebAssembly otherwise. The site itself is built with Astro and React.

Does your audio stay on your device?

Yes, for transcription. Your file is decoded and transcribed inside the browser tab, and the audio and transcript are not sent to a WhisperWeb server. Your browser still makes network requests to load the website and, on first use, to download the model files from Hugging Face; these files may then be cached. Optional anonymous usage counts stay off unless you allow them and never include audio, file names or transcript text. The privacy page explains this in detail.

Is it free?

Yes. The transcription, subtitle and conversion tools on this site are free to use, with no account and no payment. There are practical limits: currently 200 MB and 20 minutes per media file, and speed and accuracy depend on your device and recording. The terms page describes the current scope of the free tools.

Where to find the official Whisper sources

OpenAI publishes the Whisper code, model descriptions and license in its GitHub repository (github.com/openai/whisper). If you need OpenAI’s own hosted services, use OpenAI’s website and documentation. WhisperWeb is a separate, free way to run the open model in your browser.