A study on emotional feature analysis in speech signals

摘要

To achieve a comfortable speech interaction with computer, it is expected that the computer speaks with human emotion and the computer can recognize our feeling from our speech. In this paper, we try to ascertain the physical features included in our conversations. These speech signals are compared with speech signals that have no emotions by perceptual experiments and analyzing their physical features. Moreover, discrimination of new speech samples have performed by discriminant analysis. Four discriminant factors used in this analysis have generated from quantitative features of performer's speech signals. These quantitative features were defined by basing on the time variation of peak frequency of spectra. As for the emotional expression of speeches, the time variation of peak frequency of spectra and pitches were the important part.