In this question if the ML model is trained using TensorFlow (C - software) and TacoTron or TacoTron 2 (B - algorithm), which both are free for commercial use, can the owner of that creative work (A) claim copyrights on the model and its output?
Aside: I'm of the opinion that if voice actors aren't subjects to copyright claims when they do voice imitations (take Donald Duck's voice for instance), then IA voices shouldn't either.
Update: My purpose for training a ML model with some voice data set is to have some text input and make the output indistinguishable of a human voice recording. My intention is not to impersonate whoever those recordings belong to, nor hint somehow that it could be the case.
In fact, down the road I'm sure it will be a way to tweak some parameters to make the voice sound completely different of the one used to train the model.