A Chinese startup called Yunshang Qulv released an open-source text-to-speech model on July 1, 2026, that does something no commercially deployed rival currently offers: it replaces a single word ...