Classification of Emotions in Human Speech

Abstract

Dissertation thesis deals with recognition of the emotional state from human speech. The dissertation describes the current state of the Speech Emotion Recognition topic, deals with methods for speech features extraction, classification methods and is devoted to the design of a new system for speech emotion recognition. This system is modeled on the newly created emotional database emoDBova and the new database for stress detection 112DB. Designed speech emotion recognition system is implemented in secure communication infrastructure. The new databases are composed of spontaneous speech in the Czech language. The system for speech emotion recognition is designed on the basis of the last knowledge and to achieve higher accuracy than relevant proposals. The system is implemented to infrastructure, and its role is speech emotion recognition of phone call participants. Above mentioned newly created databases, a unique system for speech emotion recognition and its actual implementation in communications infrastructure are also major contributions of this work.