Abstract

This paper introduces methods used for Music Mood Classification to assist in the automated tagging of television programme theme tunes for the first time. The methods employed use a knowledge driven approach with tailored parameters extractable from the Matlab MIR Toolbox. Four new features were developed, three based on tonality and one on tempo, to enable a degree of quantified tagging, using support vector machines, employing various kernels, optimised along six mood axes. Using a “nearest neighbour” method of optimisation, a success rate in the range of 80-94% was achieved in being able to classify musical audio on a five point mood scale. This paper was originally published at The 12th International Society for Music Information Retrieval Conference, Miami, Florida (USA) October 24–28, 2011.

Parent projects