Machine-Mediated Listening (MaMeLi)

Description

MaMeLi investigates systems in which machine learning algorithms separate and enhance speech for listeners in challenging acoustic environments. The project examines how training these algorithms with room acoustic simulations of varying levels of realism affects their robustness and performance when applied to real-world data. On the user side, we study how simulated room acoustics influence speech intelligibility, listening effort, and the perceived naturalness of augmented speech. Listening tests in realistic, multisensory setups allow us to determine the required level of acoustic realism for effective machine-mediated listening in everyday situations. Our results will guide the design of future augmented hearing and speech enhancement technologies.