Guy J. Brown
Auditory Signal Processing for
The Segregation of Speech from Interfering Sounds:
A Computational Investigation of
Spatial Location and Periodicity Cues
Abstract:This paper describes two computational schemes for the segregation of speech from interfering sounds, based on auditory signal processing. In the first scheme, spectral components are grouped according to the similarity of their spatial locations. Deficiencies in this system prompted the development of a second system, in which continuity of fundamental frequency and continuity of spatial location are exploited simultaneously within a novel auditory representation, the pitch-azimuth-time cube. Results of a quantitative evaluation of the two schemes on a small data set suggest that the second system has a performance advantage over the first. Implications for the design of source segregation systems and auditory scene analysis theory are discussed.