Voice source parameter measurement based on multi-scale analysis of electroglottographic signal

This paper deals with glottal parameter measurement from electroglottographic signal (EGG). The proposed approach is based on GCI and GOI determined by the multi-scale analysis of the EGG signal. Wavelet transform of EGG signal is done with a quadratic spline function. Wavelet coefficients calculated on different dyadic scales, show modulus maxima at localized discontinuities of the EGG signal. The detected maxima and minima correspond to the so-called GOIs and GCIs. To improve the GCI and GOI localization precision, the product of wavelet transform coefficients of three successive dyadic scales, called multi-scale product (MP), is operated. This process enhances edges and reduces noise and spurious peaks. Applying the cubic root amplitude on the multi-scale product improves the detection of weak GOI maximum and avoids the GCI misses. Applied on the Keele University database, the method brings about a good detection of GCI and GOI. Based on the GCI and GOI, voicing classification, pitch frequency and open quotient measurements are processed. The proposed voicing classification approach is evaluated with additive noise. For clean signal the performance is of 96.4%, and at SNR level of 5dB, the performance is of 93%. For the fundamental frequency and the open quotient measurement, the comparison of the MP with the DEGG, Howard (3/7), the threshold (35% and 50%), and the DECOM methods show that this new proposed approach is similar to the major methods with an improvement displayed by its lowest deviation.