better training results with use_phonemes or without?

[madey83]

i have found few recipes where use_phonemes=false with characters definition:

characters_class="TTS.tts.models.vits.VitsCharacters",
pad="",
eos="",
bos="",
blank="",
characters="!¡'(),-.:;¿?abcdefghijklmnopqrstuvwxyzóąćęłńśźż‘’‚“`”„",
punctuations="!¡'(),-.:;¿? ",
phonemes=None,

and use_phonemes=True without characters

My question is: Which option give better quality results for non-english data set, like polish, slavic, etc...?