Description
Inverse problems concern the estimation of an unobserved signal from degraded observations produced by a forward process. Such problems are inherently challenging, as the inverse mapping is often ill-posed: solutions may be non-unique, unstable, or may not exist. In the audio domain, examples include stem separation, dereverberation, denoising, and timbre transfer.
Recent approaches address these challenges by learning data-driven priors with generative models. In particular, diffusion models and flow matching provide a principled framework for modeling complex audio distributions and solving inverse problems through iterative or continuous transformations.
Prerequisites
- Strong background in (statistical) signal processing
- Background in machine learning, especially deep learning and generative methods
- Interest in audio and music applications