Problems using deep generative models for probabilistic audio source separation

Maurice Frank, Maximilian Ilse

Recent advancements in deep generative modeling make it possible to learn prior distributions from complex data that subsequently can be used for Bayesian inference. However, we fi…