Research On Speech Emotion Recognition Based On Attributes Evaluation And Multi-layer Perceptron

The traditional human-computer interaction can not understand people's emotions and the humanized and intelligent human-computer interaction can improve the user's experience and comfort.As one of the most important ways of communication in human daily life,speech contains a wealth of emotional information.At present,researchers have made a lot of achievements in speech emotion recognition,but there is still a large degree in improving the accuracy of speech emotion recognition.The main work of this paper is to improve the accuracy of speech emotion recognition,it includes the following two aspects:1.Propose a speech emotion recognition model--SVM-RFE-MLP,which combines SVM-RFE and multi-layer perceptron.The extraction of high-dimensional speech emotion features will cause "Dimensional Disaster" and reduce the accuracy of speech emotion recognition,so this paper evaluates and ranks the speech emotion features based on the SVM-RFE algorithm,and then the MLP is used to classify the emotions based on the Top N ranked features.The feature subset which has the highest classification performance will be selected as the optimal feature subset.The experimental results show that the model proposed in this paper can reduce the feature dimension and improve the accuracy of speech emotion recognition.2.Proposes a speech emotion recognition method based on the spectrogram LBP texture statistical features.The spectrogram contains the voice information,this paper extractes the LBP texture statistical features of each region and the region is botained by the region segmentation of spectrogram.Then the texture statistical features of each region are cascaded and form the final texture statistical features of spectrogram which is used in the speech emotion recognition.The experiments show that the spectrogram texture statistical feature is superior to the commonly acoustic features such as short-term energy,fundamental frequency,short-time zero-crossing rate,harmonic noise ratio and so on in the classification performance of speech emotion recognition.The fusion of spectrogram features and acoustic features can effectively improve the accuracy of speech emotion recognition.