Vector Institute interview question

How would you decrease the overall latency of the system, specifically focusing on optimizing the Text-to-Speech (TTS) module within your proposed architecture?