Bayesian Modelling of Harmonic Signals for Polyphonic Music Tracking

Musical signals can be well represented as a sum of harmonically related sinusoids motivated by consideration of the sound generation mechanisms of many musical instruments. We consider the problem of estimating the harmonic model parameters in a Bayesian framework which has the potential to incorporate a priori knowledge about the structure of the data and of its parameters. Constraints can be imposed upon the time-domain variation of model parameters to reflect the high correlation of frequency and harmonic structure over time. Hierarchical model structure is represented using graphical models which allows a flexible approach to the specification of the model.