Glottal-source spectral biometry for voice characterization

The biometric signature derived from the estimation of the power spectral density singularities of a speaker's glottal source is described in the present work. This consists in the collection of peak-trough profiles found in the spectral density, as related to the biomechanics of the vocal folds. Samples of parameter estimations from a set of 100 normophonic (pathology-free) speakers are produced. Mapping the set of speaker's samples to a manifold defined by Principal Component Analysis and clustering them by k-means in terms of the most relevant principal components shows the separation of speakers by gender. This means that the proposed signature conveys relevant speaker's meta-information, which may be useful in security and forensic applications for which contextual side information is considered relevant.