Speaker’s Gender Identification based on a Biometric Study of Glottal Dynamics

P. Gómez, R. Fernández-Baillo, R. Martínez, C.Muñoz, A. Álvarez, L. M. Mazaira

Current methods in voice biometry based on an analysis of the vocal signal that directly results from lip radiation do not discriminate between articulatory features contributed by the influence and dynamism of the vocal tract, and those directly obtained from glottal dynamics. The present paper describes a biometric profile that has been obtained by a study based on the analysis of glottal dynamics. The basic signal in which the study is based is the glottal flow derivative being obtained using inverse filtering to remove vocal tract influence. The biomechanical modelling of the vocal folds allows differentiating two components in this signal: a large and slow component corresponding to the average acoustic wave (AAW) and a small and fast component corresponding to the mucosal wave correlate (MWC). The MWC has proven to be essential for voice analysis, speaker identification and the detection of pathology, as well. The analysis of the residual spectral density profile of the MWC allows extracting a biometric pattern for a given speaker. This work uses principal component analysis on this biometric pattern to show how certain parameters can be used in speaker gender characterization. A database of 100 voice samples equally balanced in gender and age have been used in the study. Results from such database in gender characterization will be shown and discussed.

Voice biometry, speaker characterization, mucosal wave, acoustic forensics

P. Gómez, R¹. Fernández-Baillo¹, R. Martínez¹, C.Muñoz¹, A. Álvarez¹, L. M. Mazaira^1
1Grupo de Informática Aplicada al Procesado de Señal e Imagen, Universidad Politécnica de Madrid, Campus de Montegancedo, s/n 28660 Boadilla del Monte, Madrid.
Tel: +34.91.3367384, fax: +34.91.3366601e-mail: pedro@pino.datsi.fi.upm.es