White Paper WHP 229 Download
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.