基于深度学习的音乐情绪分类研究

Research on Music Emotion Classification Based on Deep Learning

为解决传统音乐情感分类特征单一,导致训练效果差的问题,提出了一种多模态注意力融合网络模型,首先将执行情感分类使用的歌词和音频分离,将上下文特征提取方法与分类器相结合,从而提高特征提取效率;其次通过注意机制融合多模态特征,从而加快模型训练效率及情感分类准确率;接着提出了一种自适应孤立森林噪声方法增强模型对不均衡样本的适应性,并在一定程度上缓解模型过拟合问题.最后,将模型与LSTM、GRU、BI-LSTM、BI-GRU等模型进行仿真比较,结果表明所提模型性能最优,情绪分类准确率可达96.46%.

In order to solve the problem of poor training effect caused by the single feature of traditional music emotion classification,a multimodal attention fusion network model is proposed in this paper.Firstly,the model separates the lyrics and audio used in emotion classification,and combines the context feature extraction method with the classifier to improve the efficiency of feature extraction.Secondly,a multimodal attention fusion network architecture is proposed by integrating multimodal features through attention mechanism,so as to speed up the efficiency of model training and the accuracy of emotion classification.Then,an adaptive isolated forest noise method is proposed to enhance the adaptability of the model to unbalanced samples and alleviate the over fitting problem of the model to a certain extent.In the simulation stage,the model in this paper is compared with LSTM,GRU,BI-LSTM,BI-GRU and other models.The simulation results show that the performance of the proposed model is the best,and the accuracy of emotion classification can reach 96.46%.