ITRW on Non-Linear Speech Processing (NOLISP 07) Paris, France May 22-25, 2007

Bispectrum Mel-Frequency Cepstrum Coefficients for Robust Speaker Identification

Ufuk Ülüg (1), Tolga Esat Özkurt (2), Tayfun Akgül (1)

(1) Department of Electronics and Communications Engineering, Istanbul Technical University, Turkey
(2) Department of Electrical and Computer Engineering, University of Pittsburgh, PA, USA

In this paper, we introduce the use of bispectrum slice for mel-frequency cepstrum coefficients as robust textindependent speaker identification. The main advantage of using the bispectrum is to be able to suppress additive Gaussian noise while preserving the phase information of the signal. In order to obtain cepstral coefficients, features of the speech signal are extracted by mel-frequency filter banks, the cosine transform and the logarithm operator. Under various noisy test utterances, we compare and present the performances of the methods which use the bispectrum and the classical mel-frequency cepstrum coefficients.

Full Paper

Bibliographic reference.  Ülüg, Ufuk / Özkurt, Tolga Esat / Akgül, Tayfun (2007): "Bispectrum mel-frequency cepstrum coefficients for robust speaker identification", In NOLISP-2007, 31-34.