Review and comparison of methods using spectral characteristics for the purposes of CASA, Computational Auditory Scene Analysis
Jerry Gregoire · The Journal of the Acoustical Society of America · 2003
The brain has a remarkable ability to separate signals from a mixture. Although much progress has been made, the goal of replicating this has eluded us. Bregman (90) suggests that the brain divides this analysis into low or primitive, and high or schema processes. Schema processes require memory and subsequent interpretation. Conversely, primitive processes operate only on the innate properties of the incoming mixture and the sources that created it. The major parameters are intra aural time differences, spectrum, onset and offset, phase, and amplitude of each source. In this paper I review and compare previously proposed methods to analyze the frequency spectrum. This includes estimation of the fundamental frequency and partials produced by each source in the mixture.