Raising of the Computational Efficiency in Glottal Closure Instant Determination through Wavelet Transform*

The determination of Glottal Closure Instant (GCI) is an important task in voice signal analysis. At GCI the vocal folds close the glottis for the airstream coming from the lungs. By opening and closing the airstream passage the vocal folds generate quasi periodic impulse source for the vocal tract system. GCI determine the instantaneous voice signal pitches. They are used for speaker identification, speech recognition, voice pathology investigations, etc. GCI is of considerable importance for modelling the vocal tract too. The latter can be assumed a resonant system in free oscillating mode during closed glottis interval (CGI). In this case the vocal tract can be modeled applying linear prediction methods [1]. Some of the new methods for GCI determination use wavelet transform (WT) apparatus. It is well-known that WT are very appropriate for signal abrupt changes detection [2, 3]. Several types of wavelets have been applied for GCI determination, real [4, 5] and complex valuated too [6]. The latter determine the signal transients through transform modulus maxima and equiphase lines. The most commonly used wavelet in this case is the Morlet wavelet [3, 5, 7]. Unfortunately, its application is complicated because of large computational costs. Several algorithms for improving the computational efficiency have been proposed. The most popular among them is known as the algorithm “a trous” [8]. Another approach uses an approximation of a wavelet with close to Morlet wavelet features [9]. Novel results for GCI determination with exponentially modulated (EM) wavelet are reported in the present work. The wavelet has a close to Morlet wavelet form and allows considerable computational costs reduction without loss of GCI determination accuracy. БЪЛГАРСКА АКАДЕМИЯ НА НАУКИТЕ . BULGARIAN ACADEMY OF SCIENCES