融合说话者特征的个性化自然语音情感识别

PERSONALIZED NATURAL SPEECH EMOTION RECOGNITION BASED ON SPEAKER CHARACTERISTICS

情感特征的高级表示与说话者的个性化特征之间存在较强相关性,因此以提升个性化情感识别精度为目标,设计一组融合说话者特征和语音情感特征的识别模型,利用卷积神经网络模型获取说话者类别,在融合说话人特征高阶表达的基础上,利用卷积循环神经网络训练个性化情感识别模型,结合自建的成人自然情感语料库,在多项语音情感语料库上测试识别模型性能,从而验证该模型的有效性。

There is a strong correlation between the high-level representation of emotional features and the speaker’s personalized features. Therefore, in order to improve the accuracy of personalized emotion recognition, we design a group of recognition models which integrate speaker features and speech emotion features. The convolution neural network model was used to obtain the speaker category. Based on the high-order representation of the speaker’s features, we adopted the convolutional recurrent neural network to train the personalized emotion recognition model. Combined with the self-built adult natural emotion corpus, we tested the performance of the recognition model on multiple speech emotion corpus to verify the effectiveness of the model.