@Gates With respect to customizations for STT or TTS the customizations are limited to the supported languages of both the features. In this case it looks like Southern Hokkien is not supported.
The limits with respect to data for customization of STT or TTS these are available in the documentation but these are mostly guidance to get a good model. You can always add more data and train your model to get better customizations.
The TTS custom voice feature is the most popular feature which is used by announcers and professional voice artists to create content with voices for a specific language. These custom voices are in addition to the available standard and neural voices of azure where customers use them for various scenarios.