Hello, your model (v2) have problems with synthesis of numbers in Czech language (for example years, that are not widely used in literature) or with synthesis of Roman numerals (i.e. IV., III.). Only workaround is to run synthesis multiple times or rewrite numbers into words. There is also problem with abbreviations - it would be best if the model could just spell it, if it does not know that certain abbreviation.

Overall, your model is the best sounding TTS in Czech language i could find. Thanks!