According to your description, it should be the problem of spectrum generation.

AudioFlux itself does not directly provide TTS function, but it can be used to analyze spectrum problems, or try different spectrum types. The effect of using erb/bar type spectrum is better than mel spectrum.