Выпуск

Polishing AI Voices Without Erasing Character

My Weird Prompts · 18 ноября 2026 г. · 19 мин

Heavy speech-enhancement models often strip away vocal character, sanding down rasp and breath until distinct voices sound alike. Fixing two Chatterbox-trained AI voices required the opposite approach: a tiny equaliser cut to remove low-mid mud, followed by Resemble AI's resemble-enhance to resynthesise a missing 7.5 to 11 kilohertz band. For about a dollar of compute, the targeted two-step recipe restored clarity and top-end air while keeping the voices measurably distinct. Episode 6665 — myweirdprompts.com/n/6665 (permanent short link: myweirdprompts.com/162449) Listen on Android: MWP FM Player — myweirdprompts.com/apk

Страница выпуска