Semantic dimensionality reduction for the classification of EEG according to musical tonality

A common structural element of Western tonal music is the change of key within a melodic sequence. The present paper examines data from a set of experiments that were conducted to analyse human perception of different modulations of key. EEG recordings were taken of participants who were given melodic sequences containing changes in key of varying distances, as well as atonal sequences, with a behavioural task of identifying the change in key. Analysis of EEG involved derivation of 122120 separate dependent variables (features), including measures such as inter-electrode spectral power, coherence, and phase. The paper presents a novel method of performing semantic dimensionality reduction based on KCCA that produces a representation enabling high accuracy identification of out-ofsubject tonal verses atonal sequences.