一种基于卷积神经网络特征表征的语音情感识别方法

Feature Characterization Based on Convolution Neural Networks for Speech Emotion Recognition

为提高智能语音情感识别系统的准确性,提出了一种基于卷积神经网络CNN(Convolution Neural Network)特征表征的语音情感识别模型。该卷积模型以Lenet-5模型为基础,增加了一层卷积层和池化层,并将二维卷积核改为一维卷积核,将一维特征预处理后,输送进该模型中,对特征变换表征,最后利用SoftMax分类器实现情感分类。CASIA与EMO-DB公开数据库上的识别结果显示:与Lenet-5网络相比,所设计网络模型的准确率分别提升了1.3%与2%,与Softmax分类器相比,准确率分别提升了3.8%与6.1%。仿真结果验证了网络模型的有效性。

To improve the accuracy of intelligent speech emotion recognition system,a feature characterization based on convolution neural network(CNN)for speech emotion recognition method is proposed.a new CNN model,based on Lenet-5 model,adding a convolution layer and pooling layer,and changing the two-dimensional convolution kernel into a one-dimensional convolution kernel,is uesd to characterizing the feature.Finally,the SoftMax classifier is used to identify the types of speech emotion.The recognition results on CASIA and EMO-DB databases show that the accuracy of the designed network model increased by 1.3%and 2%respectively,compared to those of Lenet-5 network.What’s more,the accuracy rate increased by 3.8%and 6.1%respectively,compared to that of Softmax.Simulation results verify the effectiveness of the proposed model.