How we made a text-to-speech model respond in sub-50 ms

hackernews 2026-08-22 원문 보기 ↗