The goal of this paper is to explore some of the principal
spectral and statistical characteristics of music that are relevant
to the problem of Blind Source Separation (BSS). Some
BSS algorithms require some a priori knowledge of statistical
characteristics of the mixtures under examination, if
only to more accurately establish initial estimates. Furthermore,
theoretical investigations depend upon the assumption
of Wide-Sense Stationarity - the extent to which this assumption
holds is investigated.
History
Publication
Music 8th International Conference on Mathematics in Signal Processing; pp. 120-123