A Kansei model for musical chords based on the structure of the human auditory system

Consumer products are starting to be designed taking each person's tastes into account. Therefore, these products are designed based on a human sensitivity model, and by applying the human "Kansei" to various engineering fields. The paper presents a Kansei model for musical chords. The artificial neural network that realizes our Kansei model has a structure which reflects how the cochlea of the ear analyzes sound into frequencies. The input of the neural network is a musical chord consisting of three or more tones, and the output is a set of Kansei values that characterizes "cheerfulness-gloominess" "thickness-thinness" and "stability-instability." The units of the trained neural network show the following qualities: (1) Some hidden units for "cheerfulness-gloominess" and "stability-instability" are influenced by the critical bandwidth of musical tone frequencies, and furthermore these values have tonality. (2) The synaptic weights for the unit that determines the Kansei value for "thickness-thinness" becomes smaller as the tone frequencies in the musical chord become higher. The unit that outputs the weighted sum of the input frequencies mainly decides this Kansei value. These results agree with knowledge obtained from the fields of psychology and physiology.