I can assure you there is a lot of interest in the Machine Learning field to make deep fakes for music. The problem is deep fakes are not time-series based. Right now a video is treated as a bunch of independent images, whereas you couldn't do that for music/text as there are long term dependencies between time steps.
More specifically, there is a recent paper from ICLR about audio GANs (I haven't read it) https://openreview.net/forum?id=H1xQVn09FX. IMO, this tech should come out (even in app form) in the next 2-4 years.