Modeling Tonal Tension in Music Signals

Abstract

This thesis presents an exploratory study on modeling tonal tension in music signals. While the phenomenon of the tension and relaxation by harmonic event is evident to both music listeners and theorists, prediction of such perceptual harmonic tension from audio signal is still undiscovered problem. This research introduces a novel harmonic progression feature, derived from an empirically studied theoretical tonal pitch space in Western Tonal Music, and use it as a front end to train the Hidden-Markov Model (HMM) for tonal tension modeling. Using theoretical tension as a ground truth, the performance of the proposed feature is compared with other conventional harmonic features for the prediction of tonal tension in 371 songs of Bach Chorales. We also compare listeners actual response from a musical excerpt with the prediction from our model. The experimental result shows that the proposed harmonic progression feature is strongly more relevant to tonal tension than conventional harmonic features. The proposed feature is considered for a large potential incorporation into Content-based Music Information Retrieval (CBMIR). When used as a front end of feature extraction stage for mood/tension based music classification tasks, a wide-spread advantage is expected, because it reduces the complexity of chord/key detection procedures