Comparison of Spectral-tilt Features of Emotional Speech Depending on the Degree of Emotions

- Research on Emotional-speech Synthesis based on Voice-quality Conversion -

Abstract

This paper compares the voice quality features of emotions: “hot anger”, “cold anger”, “joy”, “hot sadness (crying type)”, and “cold sadness (dispirited and whispering type)” depending on the degree of the emotions in Japanese speech. Among voice-quality features, we place a focus on a spectral tilt. The analysis results show that the spectral-tilt quantities for “hot-anger” and “joy” speech are degree-dependent, i.e., the spectral-tilt quantities for these emotional speech increase significantly as the degree of the emotions becomes greater. On the other hand, the spectral-tilt quantities for “cold anger” and “cold sadness” speech do not show any significant degree-dependency, and even though those for “hot sadness” speech seem to have specific characteristics, they are speaker-dependent. These results confirm that the voice quality of “hot anger” and “joy” changes to the one whose higher-frequency band is more emphasized as the degree of the emotions increases, but that the voice-quality changes of “cold anger”, “hot sadness”, and “cold sadness” do not show any specific degree-dependency.