Abstract

IDyOM – Information Dynamics of Music (Pearce, 2005) – is a computational model of auditory expectation. It acquires knowledge of musical structure through statistical learning of sequential dependencies between notes in the music to which it is exposed. The model generates a conditional probability distribution governing the next note in a sequence, given the preceding notes. The entropy of this distribution reflects the model’s uncertainty about the next note, while the negative log probability (information content) of the note that actually follows reflects how unexpected that note is to the model. With training on an appropriate corpus of music, research using EEG , behavioural and physiological methods has shown that information content and entropy accurately predict listeners’ ongoing expectations while listening to music (Egermann et al., 2013; Hansen & Pearce, 2014; Omigie et al., 2012, 2013; Pearce, 2005; Pearce et al., 2010). The tutorial will cover: a) importing databases of music; b) extracting symbolic music features (relating to pitch, timing etc. of musical events) from the database; c) running the model to generate probabilistic descriptions of musical structure including examples using different training data, model configurations and musical features.

Marcus Pearce is a lecturer in Sound and Music Processing at Queen Mary, University of London. In addition he leads the Music Cognition Lab, is the director of the EEG Laboratory and co-director of the Centre for Mind in Society: http://webprojects.eecs.qmul.ac.uk/marcusp/