摘要
传统音乐更多的是基于歌词、韵律等单模态数据进行音乐情感鉴赏分类,难以完整的反应对象的全部特征。基于音乐特征点的非线性特征,融合音乐特征的多特征来建立一种多层神经网络结构下的前向神经网络模型,实现对音乐情感认识的分类模型。通过选择音色、力度、旋律、节奏、音高、音域、时域作为音乐特征提取变量,建立包括输入层、隐含层、输出层的前向神经网络结构模型。由输入变量完成前向传播计算,通过实际输出与期望输出间的误差梯度,引入Softmax函数进行反向激活计算,实现预定分类精度下的算法收敛。实例验证结果表明:音乐情感分类模型平均识别准确率提升到86%,在保证算法的计算效率基础上,有效地提高了算法的分类准确率。算法对节奏感、旋律和力度更强的音乐类型分类准确度更高,对于趋于平缓、轻盈的温柔情感音乐的分类准确度相对较低。
Traditional music more bases on lyrics,prosody and other single-modal data for music emotional appreciation classification.It is difficult to complete the response of all the characteristics.This paper based on the non-linear characteristics of musical feature points,establishes a forward multi-layer neural network model,to realize the classification of musical emotion cognition.By selecting timbre,intensity,melody,rhythm,pitch,range and time domain as musical feature extraction variables,we establish the forward neural network structure model including input layer,hidden layer and output layer.The Softmax function is introduced to calculate the reverse activation to achieve the convergence of the algorithm.The results show that the average recognition accuracy of the proposed music emotion classification model can be 85%,which effectively improves the classification accuracy on the basis of the computational efficiency.The algorithm has higher classification accuracy for music types with stronger rhythm,melody and intensity,and lower classification accuracy for gentle and light emotional music.
作者
宁慧
南江萍
NING Hui;NAN Jiangping(School of Humanities and Economic Management,Xi’an Institute of Traffic Engineering,Xi’an 710000,China;School of Electrical Engineering,Xi’an Institute of Traffic Engineering,Xi’an 710000,China)
出处
《微型电脑应用》
2021年第2期91-94,共4页
Microcomputer Applications
关键词
音乐鉴赏
前向神经网络
反向算法
旋律
music appreciation
forward neural network
reverse algorithm
melody
