Hi. It's of course possible to extract some *data* from voice bank files. However, unlike V, there are no traditional wave fregments in C VBs. There are HMM parameters. You won't be able to "train" a VB just using exist VBs. Like any other ML things you need original datasets. If you really interest in this (only for pure research purpose) you should explain your goal first and continue this conversitation via mail or discord.