When: Wednesday 18th November, 2015 @ 5:10 PM
Where: Room 4.31/4.33, Informatics Forum, University of Edinburgh
Modelling meter perception in music with predictive coding and perceptual inference.
Bastiaan van der Weij (University of Amsterdam)
Meter perception plays an important role in human musicality. An increasing number of studies show that meter perception is influenced by enculturation. Additionally, ethnomusicological literature has described a number of cases where individuals from different cultures have systematically different metrical interpretations of certain rhythms. Current models of rhythm perception do not account satisfactorily for effects of enculturation.
We propose an account of meter perception based on predictive coding and perceptual inference. We formally define our hypothesis, and provide a computational implementation. We model perception by activating patterns of expectation, categorised in terms of probabilistic hypotheses about candidate metrical interpretations and learned from a large collection of rhythms, proportional to how efficiently they encode the observed rhythm.
Our approach is based on an already successful model of musical expectation, which we extend to incorporate perceptual inference. Through evaluation on a corpus of artificially generated rhythms, we show that the model can, in principle, explain differences in perceived meter in terms of learning through previous exposure.
Undergratuate degree: Artificial Intelligence, University of Amsterdam
Postgraduate degree: Cognitive Science, University of Edinburgh
My master’s thesis with Mark Steedman was an attempt at using a chart-parser to parse rhythmic structure in jazz performances. Inspired by that, I applied for a PhD position at the Institute for Logic, Language and Computation back in Amsterdam, where I’m now doing a PhD supervised by Henkjan Honing and, as of recently, co-supervised by Marcus Pearce. Currently, I’m visiting Marcus at QMUL. The goal of the visit is to merge my model of metrical interpretation with his IDyOM model.