Vox — Natural Read Aloud

Vox reads web pages aloud in a voice that sounds like a person, not a phone menu. The voice is a neural text-to-speech model — Kokoro-82M — that runs entirely on your own machine. Nothing you read is uploaded, and after the one-time model download it works offline. - A moving highlight follows the voice word by word, timed from the model itself rather than estimated, so your eye can run ahead or drop back in without losing the place. - Right-click anywhere in an article to start reading from that paragraph. - Dialogue and interview transcripts are read in more than one voice, with each speaker cast separately, and headings are read as section breaks. - Every sentence is synthesised once and cached, so seeking back or reopening an article later costs nothing and the laptop stays cool. - Twelve voices, American and British. Adjustable speed. Keyboard shortcuts for play, pause, skip and stop. - A draggable scrubber and a clickable transcript for moving around a long piece. Vox has no account, no telemetry and no server. The only network request it ever makes is downloading the voice model from Hugging Face the first time you use it. Open source, GPL-3.0: https://github.com/charanteja0017/vox-read-aloud