Speech emotion recognition is one of the research hotspots in new human-computer interaction technology, and it has wide prospect in Artificial intelligence. Speech emotion recognition consists of emotional speech database's construction, feature extraction of emotional speech and speech emotion recognition classifier. This paper focuses on setting up of emotional speech database, recognizing emotional speech using LPCC, LPMCC, MFCC and ZCPA features by support vector machine, and applyingΔMFCC and MFCC_TEO to speech emotion recognition. At the same time, the composition of emotional speech database has some influences on speech emotion recognition, which has also been studied.Firstly, this paper describes how to set up emotional speech database, and construct emotional speech database with three emotions:happiness, anger and neuter, then choose appropriate sentences by subjective listening test to get the emotional speech database consisted of 890 sentences.Secondly, this paper introduces and contrasts four features:LPCC, LPMCC, MFCC and ZCPA, and recognizes the emotions using the above-mentioned features. After analyzing and comparing the experimental results, the recognition rates of the above-mentioned features are all above 80%. This paper also combines MFCC's static parameters and dynamic parameters, and transforms MFCC nonlinearly by Teager energy operator, then recognizes the emotions usingΔMFCC and MFCC_TEO. The experimental results show that the recognition rates of the improved MFCC features are better than that of MFCC.At last, the emotional speech database is divided into three kinds:English emotional speech database, Chinese emotional speech database and emotional speech database combined English and Chinese. Then LPCC, LPMCC and MFCC features are extracted from the above-mentioned emotional speech database and recognized by SVM to study the influences on the emotional recognition rate with different language speech database. There are some differences in the emotional recognition rates for different languages. The experimental results show that the emotional recognition rate of single language speech database is higher than that of speech database combined with two languages, and the emotional recognition rate of Chinese speech database is a little lower than that of English speech database. |