Glottal Closure Instant Detection from Speech Signal Using Voting Classifier and Recursive Feature Elimination

In our previous work, we introduced a classification-based method for the automatic detection of glottal closure instants (GCIs) from the speech signal and we showed it was able to perform very well on several test datasets. In this paper, we investigate whether adding more features (voiced/unvoiced, harmonic/noise, spectral etc.) and/or using an ensemble of classifiers such as a voting classifier can further improve GCI detection performance. We show that using additional features leads to a better detection accuracy; best results were obtained when recursive feature elimination was applied on the whole feature set. In addition, a voting classifier is shown to outperform other classifiers and other existing GCI detection algorithms on publicly available databases.