基于去噪自编码器的语音情感识别

Speech Emotion Recognition Based on Denoising Autoencoder

为了更加准确地进行语音情感识别,提出了一种基于去噪自编码器的语音情感识别模型。该模型用Open SMILE提取了语音中的声学特征,利用构建好的去噪自编码器获得更高阶的特征,用SVM分类器对语音中的情感进行识别分类。在EmoDB情感语料库上进行了实验,结果表明,与直接使用SVM进行分类相比,该模型对语音情感的识别准确率至少提高了2%。

In order to make the speech emotion recognition more accurate, a speech emotion recognition model based on denoising autoencoder is proposed. In this model, the OpenSMILE is used to extract the acoustic characteristics of speech, the built denoising autoencoder is used to acquire the higher-order characteristics, and SVM classifier is used for emotion recognition in speech classification. The results of experiments on the EmoDB emotional corpus show-that this model can improve the recognition accuracy of speech emotion by at least 2%, compared with the direct use of SVM.