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

Hacker News discussion: "How we made a text-to-speech model respond in sub-50 ms" — 133 points. Trending in the AI community.