Optimizing Quality vs. Latency in Real-Time Text-to-Speech AI Models | Dark Hacker News