A NEW APPROACH TO THE DETERMINATION OF THE GLOTTAL WAVEFORM

A new two-step procedure is proposed for the determination of the glottal waveform. In the first step, the inverse filter A//n(z), having zeros which correspond to the formant poles, is determined by applying the covariance method on the preemphasized speech signal s(k). Deemphasizing and integrating the residual A//n(z) S(z), the signal u(k) is obtained. In the second step, the iterative inverse filtering algorithm is applied on the signal u(k) to obtain a glottal model H//G(z). The response of the filter H//G(z) to the impulse train input having a period T (pitch period), taken in the negative time direction, defines the glottal waveform. In both steps, the algorithms are applied pitch asynchronously. The analysis of a system simulating vowel ″a,″ as well as the analysis of the five natural Serbian vowels, ″a, e, i, o, u,″ is performed. Temporal and spectral properties of the estimated glottal pulse shapes are discussed.