Machine learning for source separation, dereverberation, enhancement and source localization

DI PERSIA, LEANDRO EZEQUIEL

Oro Verde

Workshop; Workshop on Advanced Signal Processing; 2022

One of my research lines in the last 15 years has been Speech and Audio Processing. In this talk, I will present a brief review of machine learning approaches that have been applied to different problems in this area. It will cover from traditional methods like Markovian models (HMM-HMT) applied to Denoising, to representation approaches like ICA, NMF and dictionary learning, applied to Source separation, dereverberation and enhancement, to the more recent use of deep learning (LSTM and U-NETS) applied to denoising and virtual source generation/localization.