Yang Jing
Zhejiang Conservatory of Music, Hangzhou, China.
Front Psychol. 2021 Sep 28;12:760060. doi: 10.3389/fpsyg.2021.760060. eCollection 2021.
Music plays an extremely important role in people's production and life. The amount of music is growing rapidly. At the same time, the demand for music organization, classification, and retrieval is also increasing. Paying more attention to the emotional expression of creators and the psychological characteristics of music are also indispensable personalized needs of users. The existing music emotion recognition (MER) methods have the following two challenges. First, the emotional color conveyed by the first music is constantly changing with the playback of the music, and it is difficult to accurately express the ups and downs of music emotion based on the analysis of the entire music. Second, it is difficult to analyze music emotions based on the pitch, length, and intensity of the notes, which can hardly reflect the soul and connotation of music. In this paper, an improved back propagation (BP) algorithm neural network is used to analyze music data. Because the traditional BP network tends to fall into local solutions, the selection of initial weights and thresholds directly affects the training effect. This paper introduces artificial bee colony (ABC) algorithm to improve the structure of BP neural network. The output value of the ABC algorithm is used as the weight and threshold of the BP neural network. The ABC algorithm is responsible for adjusting the weights and thresholds, and feeds back the optimal weights and thresholds to the BP neural network system. BP neural network with ABC algorithm can improve the global search ability of the BP network, while reducing the probability of the BP network falling into the local optimal solution, and the convergence speed is faster. Through experiments on public music data sets, the experimental results show that compared with other comparative models, the MER method used in this paper has better recognition effect and faster recognition speed.
音乐在人们的生产生活中发挥着极其重要的作用。音乐数量正在迅速增长。与此同时，对音乐组织、分类和检索的需求也在增加。更加关注创作者的情感表达以及音乐的心理特征也是用户不可或缺的个性化需求。现有的音乐情感识别（MER）方法面临以下两个挑战。第一，第一首音乐所传达的情感色彩会随着音乐的播放而不断变化，基于对整首音乐的分析难以准确表达音乐情感的起伏。第二，基于音符的音高、时长和强度来分析音乐情感很困难，这些几乎无法反映音乐的灵魂和内涵。本文采用改进的反向传播（BP）算法神经网络来分析音乐数据。由于传统BP网络容易陷入局部解，初始权重和阈值的选择直接影响训练效果。本文引入人工蜂群（ABC）算法来改进BP神经网络的结构。将ABC算法的输出值用作BP神经网络的权重和阈值。ABC算法负责调整权重和阈值，并将最优权重和阈值反馈给BP神经网络系统。带有ABC算法的BP神经网络可以提高BP网络的全局搜索能力，同时降低BP网络陷入局部最优解的概率，且收敛速度更快。通过对公开音乐数据集进行实验，实验结果表明，与其他对比模型相比，本文所采用的MER方法具有更好的识别效果和更快的识别速度。
