Abstract
This paper presents an experimental investigation correlation between human feeling (Kansei) impression and some physical feature quantity in sound signal. As for the feature quantity, we focus on three kinds of values, that is, fluctuation value, intercept, and sum of squared errors that are obtained when making a regression analysis of sound signal in Fourier domain. In our investigation using a questionnaire survey over 34 persons, we apply multiple regression model to the relation between the feature quantity of sound signal and human evaluation about feeling impression for each person. And, we show the strength between the quantity and impression after making the construction of the regression equation for each person. Moreover, we classify the set of coefficients of the equation into three groups and discuss some tendency of human impression resulting from the quantity.
