A extraction of emotion in human speech using speech synthesize and each classifier for each emotion

Authors:
Masaki Kurematsu;Jun Hakura;Hamido Fujita
Affiliations:
Faculty of Software and Information Science, Iwate Prefectual University, Iwate, Japan;Faculty of Software and Information Science, Iwate Prefectual University, Iwate, Japan;Faculty of Software and Information Science, Iwate Prefectual University, Iwate, Japan
Venue:
ACS'07 Proceedings of the 7th Conference on 7th WSEAS International Conference on Applied Computer Science - Volume 7
Year:
2007

Citing 1
Cited 1

The production and recognition of emotions in speech: features and algorithms

International Journal of Human-Computer Studies - Application of affective computing in human—Computer interaction

Comparison of the results of factorial experiments, fractional factorial experiments, regression trees and MARS for fuel consumption data

WSEAS Transactions on Mathematics

Quantified Score

Hi-index	0.00

Visualization

Abstract

In order to estimate emotion in speech, most researchers make a classifier based on features in speech, for example power (sound pressure) and the fundamental frequency, gotten from human speech. But validity of estimation using the classifier is not good. In this paper, we discuss about how to enhance the estimation of emotion in speech. We think that making a classifier based on speech features is good for estimation. So we propose the 3 improvement points about the typical method. First point is that we use speech synthesize as training data for making a classifier. Second point is that we use not only mean and maximum but also Standard Deviation (SD), skewness and kurtosis of power and the fundamental frequency gotten from speech. Third point is that we make a classifier for each emotion and estimate emotion in speech by using them. In order to evaluate our approach, we did experiments. Experimental results show the possibility in which our approach is effective for improving typical method.