Abstract
Recent technological developments have resulted in various high quality products. Since the functionality and quality of the products are almost equivalent, the functions should be differentiated based on the sensibility thereof. Thus, we focused on push-button sounds. To estimate the quality of sounds, auditory experiments can be undertaken as a subjective evaluation, but this requires a great deal of time and a number of participants. To solve these problems, a psychoacoustic index has been proposed, although it cannot adapt to the objective evaluation of time-varying sounds such as push-button sounds. In our previous work, push-button sounds were represented in the time-frequency plane using a continuous wavelet transform, and combined with the results of the semantic differential method. From these results, the relationships between the physical properties and the auditory impressions were clarified. In this study, we propose feature extraction using triangular biorthogonal wavelets. The aim of the automated sound quality evaluation is to obtain an auditory impression score extracted by the semantic differential method without auditory experiments. Triangular biorthogonal wavelets are two-dimensional non-separable wavelets defined on a triangular lattice, allowing for multi-scale isotropic signal decomposition. The extracted features describe well the distinctive characteristics of push-button sounds. Evaluation of the automated evaluation was carried out using the sum of the squared difference and using dynamic programming matching. High recognition results were achieved with coarse features decomposed using triangular biorthogonal wavelets and dynamic programming matching. These results suggest the possibility of automated evaluation.
