There exists a body of evidence that music cognition involves statistical learning processes. Aforementioned processes relate to learning structure in sequences of musical events, informing listeners' expectations about musical events, as a piece of music unfolds in time. A successful approach to modelling musical expectation involves statistical measures of uncertainty and surprise based on Shannon's information theory.
This PhD research is concerned with evaluating the utility of information-theoretic measures of uncertainty and surprise, in the context of audio-based music content analysis. Thus, we aim to elucidate whether a cognitively inspired approach is useful in the chosen problem domain.