Evaluation of Effectiveness of Different Methods in Speaker Recognition

  • B. Šalna Forensic Science Centre of Lithuania
  • J. Kamarauskas Vilnius Gediminas Technical University

Abstract

Speaker's identification by voice is a type of biometric systems. Currently it is recent and rapidly developing science and technology with a lot of areas. Voice biometrics gives one of the worst results compared to other kinds of biometrics. The proposed new speaker's recognition method and a new system of features that consists of 36 components for that purpose. These components are formant and antiformant frequencies, their amplitudes, and various other combination parameters of spectral pairs (ratios of formant, their amplitudes and so on). Experiments with two speech databases showed that the proposed method outperformed the standard methods used for speaker identification - Gaussian mixture models using the mel scale cepstral coefficients (MFCC-GMM) and vector quantization (VQ) method. Ill. 4, bibl. 15 (in English; summaries in English, Russian and Lithuanian).

Published
2010-01-26
How to Cite
Šalna, B., & Kamarauskas, J. (2010). Evaluation of Effectiveness of Different Methods in Speaker Recognition. Elektronika Ir Elektrotechnika, 98(2), 67-70. Retrieved from http://eejournal.ktu.lt/index.php/elt/article/view/9928
Section
T 120 SYSTEM ENGINEERING, COMPUTER TECHNOLOGY