Spectral anticipations

A novel approach to (automatic) analysis of audio and music based on an "anticipation profile" is proposed. The ideas are developed from a background of information theory and basis decomposition, through the idea of a vector approach to the anticipation measurement, and ending with some higher-level reflections on its meaning for complex musical signals. The anticipation profile is estimated by evaluating the reduction in the uncertainty (entropy) of a variable resulting from prediction based on the past. Algorithms for estimation of the anticipation profile are presented, with applications for audio signal characterization and music analysis. This measure is also shown to resolve several ill-defined aspects of the structure-versus-noise problem in music and includes the listener as an integral part of structural analysis of music.