Speech Emotion Recognition Based On Spectrogram Features

At present,in the field of speech emotion recognition,the widely selected emotional characteristics of the researchers are: sound quality characteristics,frequency domain characteristics,time domain characteristics and other characteristics.However,the research on the speech frequency-domain correlation is relatively less,and the development is relatively late.In this paper,the spectrogram is used to reflect the characteristics of the speech frequencydomain correlation,and a variety of texture features are extracted,and apply the proposed feature to speech emotion recognition.The experimental results show that the recognition rate of the texture features of some of the spectrogram is relatively good,and the feasibility of the speech emotion recognition method based on the characteristics of the spectrogram is verified.In this paper,the extraction and classification of the texture features of the spectrogram are studied based on spectrogram.The main work includes the following parts:(1)Introduce the background,meaning and development of speech emotion research,as well as the commonly used methods of image texture feature extraction and texture feature classification.(2)Based on Gabor wavelet respectively combined with Gray Level Co-occurrence Matrix method,Tamura method and Local Binary Model(LBP)method,three texture feature extraction methods were used to extract the texture features of the spectrum.(3)An improved local binary model method and a fusion of LBP feature and local Hu moment feature method are proposed.Based on the emotional speech recognition in Berlin,the experimental results show that there is a certain improvement in the comprehensive recognition rate.(4)The support vector machine and the K nearest neighbor classification method are used to classify and identify the mentioned features.The recognition rate of different methods and the influence of the fusion feature weights on the recognition rate are studied emphatically.By comparing the recognition rate of the different methods,the improved LBP method and the fusion of LBP feature and local Hu moment feature method have achieved good experimental results.