If the host player A turn on this mod and using it, but another player B joins the server with this mod but doesn't turn on the python TTS engine, can the player B hear the TTS voice?

The lua mod writes player chat and voice settings to game scripts folder as a text file on the computer of the player running the mod (all players on a server run the same mods in Factorio). Each player's python script reads data from the text file and plays sound on their computer. If you do not run python script, then your mod will log text data, but it won't be converted to sound on your machine, so you will essentially not hear any voices, with no other consequences.

Thanks a lot.

It would be nice if there is a way to pass voice stream through game server , so that just the server runner needs to run the python script. (maybe?

Current version is still very good, thanks

Factorio loads sound files at startup, these cannot be loaded after. I think any way of doing this well would still require each client to have some sort of custom software running.

There is an option though: pregenerating all letters and english words. The sound files would be shipped along the mod, increasing mod size. This would mean that voices cannot be customized as there would be too many variations. This also would require slightly more UPS, as the sentences and words need to be taken apart to sounds by each client using compute time during a tick. I will try to see if this approach is realistic or sounds well enough.

So I started cooking:
https://mods.factorio.com/mod/Dynamic-voice-TTS-prerender-EN

This mod uses prerendered english voicelines to read any text without using external programs or internet.
It is not as fluid, as the sentences are glued together from separate clips at runtime, but it works out of the box.

Sounds good enough to me!