Glottal Closure Instant Detection by the Multi-scale Product of the Derivative Glottal Waveform Signal

Abstract

This paper is about the detection of the glottal closure instants (GCI) by the multi-scale product (MP) of the derivative glottal waveform signal. Based on the source filter model, the derivative glottal waveform signal is estimated by the inverse filtering of the non pre-emphasized speech signal with the LP coefficients. The derivative glottal waveform signal represents the real excitation of the vocal tract and shows discontinuities at GCI. MP acts as a discontinuity detector. A preprocessing step is added to improve the GCI detection. The performance of our method is evaluated on the Keele university database and compared to the MP applied directly on the speech signal. Using the preprocessing phase, the MP applied on the derivative glottal waveform signal gives an identification rate of 99.21 % and an accuracy to \(\pm \)0.25 ms of 87.32 % versus an identification rate of 99.15 % and an accuracy to \(\pm \)0.25 ms of 75.78 % for the MP method applied directly on speech signal.