How we made a text-to-speech model respond in sub-50 ms Published 2026-08-22 · news How we made a text-to-speech model respond in sub-50 ms