An Auditory Scene Analysis Approach to Speech Segregation

This proposal seeks to investigate monaural (one-microphone) speech segregation. Motivated by the observation that the human auditory system has the remarkable ability of segregating an acoustic mixture and grouping the components of the same event into a stream in a perceptual process called auditory scene analysis, the proposed approach is based on auditory scene analysis principles. The overall goal of the proposed research is to develop a speech segregation system that substantially improves speech intelligibility of listeners in noisy environments. To achieve this goal, the proposed segregation system is designed to have several stages of computation, including early processing, simultaneous organization and sequential organization. The proposed system will be evaluated using mixtures of speech and a variety of interfering sounds, and the success of the system will be measured in terms of signal-to-noise ratio and speech intelligibility in noise. BENEFIT: The results of this project are expected to significantly improve the speech intelligibility of listeners in real-world acoustic environments, which in turn could lay the technical foundation for a wide range of applications. Commercial potential includes better hearing aid and cochlear implant design to benefit a large population of people with varying degrees of hearing loss, robust automatic speech and speaker recognition, and human speech communication in adverse civilian and military environments.