The speech model downloads to your browser the first time you use it (once cached, later visits skip the download). Everything runs on your own device — the file never leaves your computer. Works best in Chrome or Edge. Longer files take longer to process, especially without a GPU-capable browser.