Problems using deep generative models for probabilistic audio source separation
This identifies a critical limitation in applying deep generative models to audio source separation, which is incremental as it highlights specific issues rather than proposing a new solution.
The paper found that deep generative models for audio produce prior distributions that are either too peaked or too smooth, making them unsuitable for probabilistic audio source separation, as quantified on two models and datasets.
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 find that distributions learned by deep generative models for audio signals do not exhibit the right properties that are necessary for tasks like audio source separation using a probabilistic approach. We observe that the learned prior distributions are either discriminative and extremely peaked or smooth and non-discriminative. We quantify this behavior for two types of deep generative models on two audio datasets.