Whisper

Main introduction
Whisper.fans is a lightweight, browser-based front-end for OpenAI’s Whisper automatic-speech-recognition model. The site lets users upload any audio or video file, choose one of the five Whisper model sizes, and receive a time-stamped transcript in minutes without installing software or creating an account.

Key features extracted from the homepage
– 100 % client-side processing – audio never leaves the user’s device, guaranteeing privacy.
– No registration, no usage limits, no ads; the page is open-source under MIT license.
– Supports every file that the browser can decode (mp3, mp4, wav, m4a, flac, ogg, webm, mov, etc.).
– Offers the full Whisper model family (tiny → base → small → medium → large) so users can trade speed for accuracy.
– Outputs plain text, VTT or SRT subtitles with optional word-level timestamps; transcripts can be edited on-page and copied or downloaded as .txt, .vtt or .srt.
– Runs in any modern Chromium, Firefox or Safari desktop browser via WebAssembly and GPU-accelerated ONNX runtime; no server backend, so it works offline after first load.
– Page weight < 3 MB; model binaries are cached in IndexedDB, avoiding repeated downloads.