Perceptual aspects of synthesized approximations to melody.

Abstract

A procedure is described for synthesizing tone sequences that contain a high proportion of events that are likely to be found in melodies that occur in music. This is done by using a random-number generator with probability weightings that favor the selection of certain notes and intervals. The form of the probability weightings derives from constraints on the "semitone span" of the intervals, the "fifth span" of the intervals, and the occurrence of "scale" notes. The role of "redundancy" is also considered. In this way it is possible to obtain melodies that obey various combinations of the different constraints. Other melody variables are either fixed, randomized, or controlled. These experimental melodies were used in a test of the hypothesis that the more melodious tone sequences will be easier to organize perceptually and remember than less melodious sequences of equal redundancy. A psychophysical scaling procedure confirmed that the constraints generated tone sequences bearing degrees of perceptual similarity to "real" melodies. Melody discrimination was also measured. The "fifth span" and "scale" constraints increased the melodiousness of tone sequences. They also improved performance on tasks requiring the discrimination of unfamiliar, untransposed melodies, and the discrimination of transposed melodies after some familiarization. It is argued that general perceptual principles do not give a complete account of the perception of melody: A full description requires consideration of the listener's tacit musical "knowledge" and its interaction with perceptual processes.