Automatic detection of tonality using note distribution

Western musical compositions written in the period of common-practice tonality tend to use key signatures which are effectively a measure of their local harmonic ‘weight’. The authors report a rule-based system for the automated detection of tonality which makes use of the total duration of individual note classes. From a collection of experimental data, the distribution of note classes is observed. Using these empirical results a heuristic search is made with rules derived from high-level musical knowledge in order to estimate the tonality. The resulting data can be used to assist the recognition of partially detected key signatures in an Optical Music Recognition system.