This paper deals with the evaluation and the comparison of two methods for determining glottal closure instants (GCIs) from the speech signal. These methods are the Multi-scale Product Method (MPM) and the Dynamic Programming Phase Slope Algorithm (DYPSA). The MPM consists of determining abscissa of peaks appeared on the product of wavelet transform coefficients at three adjacent dyadic scales. The wavelet used to compute the product is the quadratic spline function. However, the DYPSA algorithm is based on the group delay function to generate the GCI candidates and the dynamic programming technique to select the true GCIs. The comparison of the two algorithms is operated without noise and with additive Gaussian noise at various SNR levels. The evaluation is operated on the Keele University database. Results show that the MPM is more robust to noise than the DYPSA with a good detection rate nearby 90% at an SNR level of 0 dB and a small false alarm rate. Besides, the DYPSA is more sensitive to noise with a good detection rate of 20% at 0 dB and a high false alarm rate surpassing 70%.
Speech, Glottal closure instant detection, DYPSA algorithm, Multi-scale product, Noise
