EMOTIONAL QUALITY EVALUATION FOR GENERATED MUSIC BASED ON EMOTION RECOGNITION MODEL

Hongfei Wang,Wei Zhong,Lin Ma,Long Ye,Qin Zhang

DOI: https://doi.org/10.1109/icmew56448.2022.9859459

2022-01-01

Abstract:In the field of musical emotion evaluation, the existing methods usually use subjective experiments, which are demanding on the experimental environment and lack of unified evaluation standard. This paper proposes an emotional quality evaluation method for generated music from the perspective of music emotion recognition. In the proposed method, we analyze the correlation between audio features and emotion category of music, and choose MFCC and Mel spectrum as the most significant audio features. And then the emotion recognition model is constructed based on residual convolutional network to predict the emotion category of generated music. In the experiments, we apply the proposed model to evaluate the emotional quality of generated music. The experimental results show that our model can achieve higher recognition accuracy and thus exhibits strong reliability for the objective emotional quality evaluation of generated music.