Triple Comparison-based Interactive Differential Evolution for Creating Sign Sound

Interactive Evolutionary Computation is known as effective method for creating media contents suited to each user in the fields of graphics and sounds. In the IEC tasks, human users' fatigue is one of the serious problems in terms of the number of searches is limited. The present study tackles this problem by employing a subjective comparison of three solution candidates. Triple comparison-based Interactive Differential Evolution (IDE) was proposed by Pei et al., and this study proposes a method of creating sound with the IDE. To investigate efficiencies of the IDE, listening experiment was conducted. Eleven subjects participated in the experiment. In search step, the subjects compared three sign sounds and selected the best one with a perspective of "warning". In evaluation step, the subjects scored the two representative sounds which were picked up from the initial and the final generations. Distance between target vectors, the number of selections of each triple vector, and subjective fitness value were used as indices for investigating performance of the IDE. Decrease in the distance was observed in the prior half of the generations, and decrease in the selections of non-target vectors was observed. The subjective fitness increased, however, it was not significant. The performance of the triple comparison-based IDE for creating warning sound was discussed with good and poor experimental results, and some points should be improved were suggested in this paper.