Speech generation and voice cloning run locally in your browser.
Use a clean 6–12 second voice sample with no music or background noise.