Multitaper Estimation of Frequency-Warped Cepstra With Application to Speaker Verification

Johan Sandberg (1) , Maria Hansson-Sandsten (1) , Tomi Kinnunen, (2) , Rahim Saeidi (2) , Patrick Flandrin (3) , Pierre Borgnat (3)

1 Mathematical Statistics, Centre for Mathematical Sciences
2 Speech and Image Processing Unit, Department of Computer Science and Statistics
3 Phys-ENS - Laboratoire de Physique de l'ENS Lyon

Usually the mel-frequency cepstral coefficients are estimated either from a periodogram or from a windowed periodogram. We state a general estimator which also includes multitaper estimators.We propose approximations of the variance and bias of the estimate of each coefficient. By using Monte Carlo computations, we demonstrate that the approximations are accurate. Using the proposed formulas, the peak matched multitaper estimator is shown to have low mean square error (squared bias + variance) on speech-like processes. It is also shown to perform slightly better in the NIST 2006 speaker verification task as compared to the Hamming window conventionally used in this context.

Keywords: Cepstral analysis; MFCC; Multiple windows; Multitapers; speaker verification; Speech analysis

Domains: Computer Science [cs]; Signal and Image Processing; Engineering Sciences [physics]; Signal and Image processing

Johan Sandberg, Maria Hansson-Sandsten, Tomi Kinnunen,, Rahim Saeidi, Patrick Flandrin, et al.. Multitaper Estimation of Frequency-Warped Cepstra With Application to Speaker Verification. IEEE Signal Processing Letters, 2010, 17 (4), pp.343-346. ⟨10.1109/LSP.2010.2040228⟩. ⟨ensl-00475928⟩