Research On Speech Emotion Recognition Algorithm

With the development of computer science,speech signal processing is widely used in all aspects of society.At present,speech emotion recognition technology has become the key of human-computer interaction system.To make human-computer interaction more convenient and humanized,researchers began to study the emotional signals of speech.By analyzing the emotions of the operators,the intelligent human-computer interaction system can be more active and accurate to achieve the requirements of the operators,and can timely adjust the form of the dialogue,make the communication more intelligent.The main work of this paper is as follows.(1)The speech emotion characteristic parameters were optimized.After the optimization and integration of the frequency cepstrum coefficient,MFCC is combined with prosodic features and sound quality features as the characteristic parameters of speech emotion recognition.The experimental results show that the new MFCC coefficients obtained from the mixture of MFCC,I-MFCC and Mid-MFCC have significantly improved the identification ability of the entire frequency band.(2)The speech emotion recognition algorithm based on F-MFCC parameters is proposed in this thesie.Fisher ratio criterion was used to combine MFCC and its derived parameters I-MFCC and Mid-MFCC to generate F-MFCC.After the generated F-MFCC is mixed with other feature parameters,the speech emotion recognition is performed by using different spectral-based feature parameters.The experimental results show that using F-MFCC as the characteristic parameter can further improve the recognition rate of the recognition model and reduce the dimension of the feature parameters to some extent.(3)A speech emotion recognition method based on the new decision model is proposed.The recognition results obtained by the BP neural network,the support vector machine and the K-nearest neighbor algorithm are passed through a voter,and the output of the voter is used as the recognition result.Experimental results show that the new decision model can reduce the probability of speech misjudgment and further improve the average recognition rate of the final speech emotion.