An Auditory Scene Analysis Approach to Speech Segregation

ABSTRACT: This Phase II proposal seeks to develop a prototype speech segregation system that can significantly improve intelligibility of noisy speech for human listeners, a technical challenge that has eluded a solution for decades. Motivated by principles of auditory perception, system development is based on an auditory scene analysis approach to speech segregation. This approach differs substantially from traditional speech enhancement techniques, and its promise has been successfully demonstrated in Phase I research. The proposed effort will address both simultaneous organization and sequential organization in computational auditory scene analysis, as well as efficient implementation. The prototype system will be systematically evaluated using a set of clear performance milestones involving normal-hearing and hearing-impaired listeners and a variety of speech and background interference. BENEFIT: Speech segregation technology that can improve speech intelligibility in noisy environments for human listeners has huge potential for commercialization. Practical applications include hearing aid design and cochlear implants where noise reduction is a major unsolved problem, better consumer devices for speech communication in adverse acoustic conditions, and robust automatic speech and speaker recognition. These applications occur in both civilian and military domains.