데이터 증강에 따른 스팩트로그램 기반 차량 주행음의 감성적인 음질 특성 분류 정확도 변화

Along with the development of the technology of electric vehicles using a motor instead of an internal combustion engine, research on the active acoustic design for generating a vehicle driving sound that is desired by users is being actively conducted. Therefore, the need to quantitatively evaluate effective sound quality characteristics of a vehicle according to existing expert judgment, through a virtual model such as a deep learning-based algorithm has increased. Deep learning-based algorithms can improve the accuracy of prediction when there is a large amount of learning data; however, obtaining such data in actual industrial sites is difficult. In this study, we presented a change in the accuracy of a deep learning-based algorithm based on the number of data augmentation. Additionally, we verified the basis of the algorithm's judgment by implementing the explainable artificial intelligence technique.