Automatic blind source separation of speech sources in an auditory scene

摘要

Hearing-impaired individuals have much more difficulty, then normal hearing individuals, with speech intelligibility in cocktail party situations. Hence, the overall research goal is to improve intelligibility of speech and to enhance spatial hearing of hearing aids in these situations. It is known that spatial audio leads to improved speech intelligibility. As a step towards this goal, source separation algorithms were investigated. In particular, the DUET Blind Source Separation (BSS) algorithm was modified to automatically extract speech sources in a simulated auditory scene.