Extraction of glottal features for speaker recognition
Stevan Ostrogonac, Milan Sečujski, Dragan Knezevic, Siniša Suzić · 2013
This paper presents an extension to the SEDREAMS algorithm for extracting the information about glottal opening and glottal closure instants (GCI and GOI) directly from the speech signal. Accurate detection of GCIs and GOIs is crucial for estimating the glottal features which are to be used in speaker recognition systems. Many different approaches resulted in a variety of algorithms dealing with this problem. The algorithm that showed the best results so far consists of two steps. First, a mean-based signal is computed to determine the intervals in which the GC! and GOI moments should be searched for. Then, discontinuities are sought in the LP residual of the speech signal and they represent the estimation of glottal events. This algorithm (in literature found under the name SEDREAMS) is widely used in glottal excitation estimation systems. However, the mean-based signal calculated in the first step of the process sometimes contains unwanted spectral components which significantly degrade the performances. This paper describes one way to address this problem. By applying an adaptive filter to the mean-based signal significant improvement has been achieved in glottal features estimation. This was confirmed by a speaker recognition experiment which showed very encouraging results.