Glottal Source Cepstrum Coefficients Applied to NIST SRE 2010
Luis Miguel Mazaira Fernández, Agustín Álvarez-Marquina, Pedro Gómez‐Vilda, R. Martínez, Cristina Muñoz · 2010
Through the present paper, a novel feature set for speaker recognition based on glottal estimate information is presented. An iterative algorithm is used to derive the vocal tract and glottal source estimations from speech signal. In order to test the importance of glottal source information in speaker characterization, the novel feature set has been tested in the 2010 NIST Speaker Recognition Evaluation (NIST SRE10). The proposed system uses glottal estimate parameter templates and classical cepstral information to build a model for each speaker involved in the recognition process. ALIZE [1] open-source software has been used to create the GMM models for both background and target speakers. Compared to using mel-frequency cepstrum coefficients (MFCC), the misclassification rate for the NIST SRE 2010 reduced from 29.43% to 27.15% when glottal source features are used