Making NeMo TTS faster on CPU

[NoteToSelfFindGoodNickname]

Hello!

I would like to use NeMo TTS as a screen reader.

NeMo TTS is already faster than Coqui, but synthesizing still takes too much time for a screen reader.
For 1 sentence, it needs about 1000 ms on a good CPU.

Blind people however expect the speech output to happen instantly even on not-so-good devices.
So using a GPU or a really good CPU is not an option here.

Does anybody know how to make NeMo TTS synthesizing faster than it is right now on CPU?

Thank you!