Natural neural voices, generated entirely on your device — nothing you paste ever leaves your browser. First use downloads the voice model (about 90–330 MB depending on your device); after that it’s cached and instant.