A Pitch-Dependent Glottal Source Model for Artificial Voice

An experimental study is reported which provides information to improve the modelling of the glottal source for the synthesis of natural sounding voiced speech by means of an artificial larynx. Fourteen healthy New Zealanders (6 males and 8 females) with English as their native language were required to utter the target vowels (/a/, /e/, /i/, /o/, /u/ and /er/) at the speaking range and at the singing range. Acoustic and electroglottographic (EGG) signals were simultaneously recorded during the utterances. A new model known as the “twin-bar model” is presented. It is based on the opening phase (OP), closing phase (CP), steepest slope at the closing phase (sCP) and fundamental frequency (F0) of the EGG signals with the first three variables dependent on F0. Results from the experiment show that ln OP, ln CP and sCP are all negatively related to ln F0. The twin-bar model offers a simple single parameter glottal source that varies with F0