> Our audio research framework and training code is released under the MIT license to enable the broader community to reproduce and build on top of our work
It's pretty common in academic research for trained model weights to be licensed under something different from the code that one would run to create such a model if one had both sufficient compute resources and the same training dataset. That is, if those weights are ever released at all!
IMO, while I'd rather have one part permissively licensed than nothing at all... it stinks that companies sponsoring researchers get an un-nuanced level of street cred for "open sourcing" something that they know nobody will ever be able to reproduce because their data set and/or their compute grid's optimizations are proprietary.
As it stands, I'm not at all sure that the outputs of this model can be used for commercial videos.