Pendrokar/xVASynth-TTS · xVASynth is poor in foreign languages?

If you do ask this question. Then yeah, the multilingual capabilities were never properly tested. I personally blame it on NOT dropping ARPAbet for IPA when going from v2 to v3 (xvapitch).

Instead espeak and perhaps other phonemizers were used to convert text to IPA, and then that IPA got converted to ARPAbet. With extra ARPAbet symbols added, which I in the Open TTS table called ARPAbet+.

IMHO, base model has to be recreated with IPA. It would make it incompatible with v3, so that would mean it would be a v4 model.