Skip to main content

Multichannel Modeling for Enhancing and Masking Speech Communications

Researchers: Ted Selker, Ian Lane

Research Area: Mobility

Cross Cutting Thrusts: Usable Privacy and Security

Abstract

Scope: Humans have a remarkable ability to comprehend speech even in the presence of multiple simultaneous speakers; this is known as the “cocktail party effect”. In prior work we demonstrated that this effect goes away when certain acoustic cues are not present. In this project, we propose to explore how stereo positioning of voices and the acoustic characteristics of voices can allow cocktail party effects in a computer manufactured audio landscape. We will investigate the use of this technology in acoustic environments for multiparty teleconferencing and also evaluate its feasibility to inhibit external eavesdropping.

Outcomes: Novel modeling approaches for controlling multichannel acoustic landscapes. Algorithms for dynamic morphing and generation of speech signals in multichannel landscapes. Empirical evaluations based on human comprehension