基于前向神经网络的音乐情感识别分类算法

Music emotion recognition classification algorithm based on forward neural network

为解决单模态数据在音乐情感分类上的局限性,并同时提高对音乐情感分类的准确性,文中提出了一种基于前向神经网络的多特征融合音乐分类算法。在传统的前向神经网络模型中融入切比雪夫正交多项式簇作为隐藏层各神经元的激励函数,使每一层神经元的激励函数各不相同。利用梯度下降学习算法来进行网络参数的有监督训练;同时利用音频、歌词中不同模态的数据,使其形成多模态数据,来进行音乐情感分类模型的训练。实验测试结果表明,该算法对音乐情感的分类具有较好的效果,平均准确率为78.37%,具有良好的有效性与可行性。

In order to solve the limitation of single-modality data in music emotion classification and improve the accuracy of music emotion classification,a multi-feature fusion music classification algorithm based on forward neural network is proposed in this paper.In the traditional forward neural network model,the Chebyshev orthogonal polynomial cluster is incorporated as the excitation function of each neuron in the hidden layer,so that the excitation functions of each layer of neurons are different.Gradient descent learning algorithm is used to supervise training of network parameters.At the same time,the multi-modal data is formed by using different modal data in audio and lyrics to train the music emotion classification model.The experimental results show that the algorithm has a good effect on the classification of musical emotions,and the average accuracy is 78.37%,which has good validity and feasibility.