A Proposal of Emotional Detection System from Speech Data

摘要

In this paper, prosodic characteristics are obtained from 4 kinds of emotional speeches that are neutral, angry, sad, and joyful speech, and are analyzed. Furthermore, each emotional speecn is classified by using a neural network. In prosodic characteristics, the pitch has characteristics of frequency domain and the length of utterance time has characteristics of time domain. These elements are considered to be the most important elements. A prosodic characteristics extraction method is employed a special integrated circuits (IC) , which combine analog and digital processing. We have already gotten a patent for this IC.