About

Most "free subtitle generator" sites ask you to upload a video to their server before they give anything back. That means your raw footage, your interview, your unreleased podcast sits on someone else's machine — and it means quotas, queues and sign-up walls.

Why in the browser

Modern browsers can run a real speech model locally via WebAssembly. If the compute happens on your device, there is nothing to upload and nothing to meter: no queue, no daily limit, no length limit, and it keeps working without a connection once the model is cached.

What this means in practice

• Your media never leaves your computer.
• No account, no email, no quota.
• Processing is slower than a big cloud GPU — roughly 0.6× realtime on a desktop.
• So the tool is aimed at clips of a few minutes rather than hour-long recordings.

Under the hood

Speech recognition runs on Whisper-base (quantised to about 75 MB) through transformers.js and ONNX Runtime Web, single-threaded in WebAssembly. Single-threaded means no cross-origin isolation headers are needed, which keeps the page compatible with everything else a normal website needs.