An Exploration Of Speech Emotion In Different Languages Based On Machine Learning

With the development of globalization,the communication between different languages becomes more and more frequent,and multi languages learning becomes more and more important.The understanding of language is not only about the understanding of semantic text,but also the difference of emotion.Therefore,the understanding of different language speech emotion can help different language learners better.For cross language learners,by directly understanding the emotional differences between the two languages,they can master the new language more quickly and conveniently.With the continuous development of artificial intelligence technology,speech emotion recognition technology has been continuously developed.The computer can automatically extract the corresponding acoustic features according to the input speech data,which lays a foundation for the exploration of speech emotion in different languages.Compared with the traditional speech recognition method,machine learning method is more objective and efficient,and it is suitable for experiments with large amount of data.In this paper,machine learning method is used to classify the speech emotion databases of different cultures,and the similarity and difference of speech emotion expression of different cultures are compared and studied.The main research work of this paper is as follows:1、Three kinds of classifiers,support vector machine,convolution neural network and long-term memory network,are used to build a speech emotion recognition model,which has a good recognition rate.The eGeMPAS acoustic feature set is adopted,including 88 dimensional acoustic features,mainly including loudness,formant,Mel cepstrum coefficient and other acoustic parameters.It avoids the waste of resources caused by the feature set with large amount of data.2、In the comparative study,we adopt the method of transfer learning,choose a corpus as the training set,establish the speech emotion classification model,choose other corpus as the test set to get the classification results,so as to compare the difference and similarity of the speech emotion between the two corpora.Experiments show that there is a high similarity in the expression of sad emotion,while the difference in the expression of happy emotion is the largest.Compared with English and Chinese,German has the most obvious expression and the highest recognition rate.The neutral emotion in English and the sad emotion in Chinese have better generalization,and can recognize the voice emotion of different cultures.The sad emotion in English and the neutral emotion in Chinese have better adaptability and can adapt to different models.The generalization and adaptability of German emotion are poor.