Characterizations of the voice spectrum in emotional utterances like the Hammarberg index or the amount of energy below a certain pivot point often ignore speaker dependent pitch information. Furthermore, these approximations of spectral slope are not very sensitive to details of the spectral shape. Here, we present three approaches to incorporate speaker specific information in determining the appropriate pivot value for distinguishing between low and high frequencies. Additionally, three different approaches of characterizing the spectral slope of the LTAS (Long-Term Average Spectrum) are presented. The validity of these approaches is tested on a corpus of emotional speech developed at the University of Geneva. The results show the feasibility and usefulness of taking speaker specific information into account as well as more extensive characterizations of the slope of the LTAS.
Cite as: Tamarit, L., Goudbeek, M., Scherer, K. (2008) Spectral slope measurements in emotionally expressive speech. Proc. ISCA ITRW on Speech Analysis and Processing for Knowledge Discovery, paper 007
@inproceedings{tamarit08_spkd,
author={Lucas Tamarit and Martijn Goudbeek and Klaus Scherer},
title={{Spectral slope measurements in emotionally expressive speech}},
year=2008,
booktitle={Proc. ISCA ITRW on Speech Analysis and Processing for Knowledge Discovery},
pages={paper 007}
}