The Establishment Of Mandarin Speech Emotion Acoustic Characteristic Database

In order to strengthen the intelligence and humanity of computer, develop a newenvironment of human-machine and inrease the accuracy of a speech recognitionsystem, it is very important for us to study speech emotion recognition. The aim of thispaper is to analyze the relationship between acoustic features and emotion of voicesignal, and establish the database of speech emotion recognition, and make arrangementfor questing a new method of speech emotion recognition.Combining with the Application research project of Sichuan Province named by"speech emotion recognition based on rough set (2008RQ002)", speech emotionrecognition is studied in this paper. Furthermore, we elaborate recent developments ofspeech emotion recognition, recognition method, and emotional speech database in thedomestic and foreign, analyze acoustic feature of voice signal and its extraction method,discuss emotional feature analysis, emotion recognition of speech and emotionmodeling. In particular, we set up emotional speech database by30acoustic features.The main contents of the thesis are as follows: as the study of Mandarin’s affectivespeech is starting stage, there are less technologies, experiences and methods aboutemotional speech database. For the need of study, we establish a Mandarin’s affectivespeech database, and by using recording method, we collect1600emotional sentences,including six emotional categories which are anger, fear, joy, surprised, sadness andcalm. At the same time, many experts are invited to listen all these sentences. Afterlistening these sentences, it removes549sentences which include ambiguous affectionstate or semantics is not clearly. Finally, the remainder1051sentences are effective.Every speech sentence has been processed by using software, such as cool edit pro v2.1and praat etc. These soft wares process earlier datas of acoustic features. And then, ituses BP neural network to simplify all acoustic features from software.Throughextracting effective features from earlier model, it obtain30effective acostic features.So it can obtain the acoustic characteristic data of the speech. It shapes a datasheet of1051*30. On the base of datasheet, it establishs database on the SQL Server2000 system. The database stores the acoustic characteristics of each emotional speech, suchas frequency, energy, duration and resonant etc. It provides a good experimental dataplatform for the study of Mandarin’s emotional speech identification.