SEMINAR:Sound separation: an overview

Speaker: Hakan Erdoğan

Title: Sound separation: an overview

Date/Time:11 May 2022 / 18:00 - 19:00

Zoom link: https://sabanciuniv.zoom.us/j/4945350538

Abstract:In this talk, the problem of sound separation from single channel recordings will be described and solutions will be discussed. We will briefly go over the methods and algorithms used for this problem in the pre-deep-learning era. Some of the classical methods include statistical methods, non-negative matrix factorization, and independent component analysis. Currently, deep learning based architectures are achieving state-of-the-art results in sound separation. Favorite architectures, loss functions and estimation styles that work well for these problems will be covered. Some recent efforts towards unsupervised or self-supervised training of sound separation models will be motivated and described. Finally, possible future directions will be discussed.

Bio: Hakan Erdogan received his Ph.D. from University of Michigan. He previously worked at IBM, Sabanci University, MERL and Microsoft. He is currently a research scientist at Google. His current research interests are in using deep learning and signal processing for sound separation.