SciELO - Scientific Electronic Library Online

vol.12 issue3Integrated navigation algorithm for autonomous vehicles with low cost technologyWiener filtering to noise reduction for speaker verification. author indexsubject indexarticles search
Home Pagealphabetic serial listing  

Services on Demand




  • Have no cited articlesCited by SciELO

Related links

  • Have no similar articlesSimilars in SciELO


Revista Cubana de Ciencias Informáticas

On-line version ISSN 2227-1899


CAMPBELL HERNANDEZ, Edward L; HERNANDEZ SIERRA, Gabriel  and  CALVO DE LARA, José R.. Method of robust feature extraction for a diarization system.. Rev cuba cienc informat [online]. 2018, vol.12, n.3, pp.140-151. ISSN 2227-1899.

Automatic Speakers Recognition Systems are biometric systems that allow the identification and verification of people, using voice as a discriminatory feature. One of the challenges to overcome during the recognition process is when the audio flow to be processed has several speakers, since it’s necessary to have knowledge of the temporal location of the audio segments relative to each speaker, in order to be able to directly compare those segments with the speaker samples stored in the enrollment database. The diarization system allow to define the audio regions that are associated to a same speaker, solving, the mentioned problem in the recognition process. In this article is proposes a robust feature extraction technique as subsystem of the diarization system, called Perceptive Minimum Variance Distortionless Response, which demonstrated greater robustness to noise than the dominant technique in state-of-the-art, Mel Frequency Cepstral Coefficients. Experimentally is demostrated as the feature proposed present a level less of variance compared with the mel feature, between clean and noisy frame, subjecting the audio to a signal noisy relation of 6 dB and 8 dB respectively.

Keywords : diarization; perceptive minimum variance distortionless response; robust feature.

        · abstract in Spanish     · text in Spanish     · Spanish ( pdf )


Creative Commons License All the contents of this journal, except where otherwise noted, is licensed under a Creative Commons Attribution License