Abstract

It is well known that male and female speech is different. Several studies investigated differences in articulation, phonation, resonance and voice characteristics. According to the source-filter model of speech production, the filter - the vocal tract - carries the main information about the identity of the speaker. Hence the excitation signal was often neglected in speech analysis and synthesis. However, additional research about the excitation signal showed differences in male and female speech. For instance the contact quotient, which describes the vocal fold movements, had a significant greater mean for male than for female speech.

In this Master’s thesis electroglottographic recordings obtained from female and male speakers from three different databases were analyzed with regard to gender differences. Therefore, jitter, shimmer and open quotient computations were performed. In contrast to many studies which used recordings of prolonged vowels, data used in this thesis also consisted of recordings of read German sentences which is more natural than sustained vowel phonation. It was shown that results in earlier studies regarding the open quotient obtained from prolonged vowel and repeated syllable analysis are valid for read speech as well. Using the glottal time instants found in the electroglottographic signal, an excitation signal was synthesized using the LF-model for the glottal flow. Using this method of excitation signal synthesis, an individual excitation signal was generated. The characteristics of this signal contributes to gender perception in addition to the fundamental frequency value.

Contact:

Martin Hagmüller ({hagmueller}@tugraz.at or 0316/873 4377) Anna Katharina Fuchs ({anna.fuchs}@tugraz.at or 0316/873 4367)