Method for the Maximization of the Likelihood Function of Speech Autoregressive Parameters Based on Line Spectrum Pairs
The paper considers the estimation of the parameters of autoregressive model at additive white noise background. The principle of maximum likelihood is used for this purpose. The main goal is to find the maximum of likelihood function depending on parameters of autoregressive model. Representation of likelihood function through line spectrum pairs and other alternative parameters is presented. This provided possibility of likelihood function maximization by KNITRO algorithm. The presence of multiple local minima of the considered likelihood function is shown. Experimental results including the comparison with widely used expectation-maximization (EM) method are presented for the real speech signals.
Повний текст:PDF (English)
Rabiner L. R. Theory and Applications of Digital Speech Processing / L. R. Rabiner, R. W. Schafer. — N.-J. : Prentice-Hall, 2011.
Gannot S. Iterative and sequential Kalman filter-based speech enhancement algorithms / S. Gannot, D. Burnstein, E. Weinstein // Transactions Speech Audio Processing. — 1998. — Vol. 6. — P. 373–385.
Semenov V. A novel approach to calculation of line spectral frequencies based on inter-frame ordering property / V. Semenov // Proc. IEEE Conf. ICASSP. — 1998. — Vol. 6. — P. 1072–1075.
Domes F. The optimization test environment / F. Domes, M. Fuchs, H. Schichl, A. Neumaier // Optimization and Engineering. — 2014. — Vol. 15. — P. 443–468.
Byrd R. Large-Scale Nonlinear Optimization, chapter KNITRO: An Integrated Package for Nonlinear Optimization / R. Byrd, J. Nocedal, R. Waltz. — Springer, 2006. — P. 35–59.
Kalyuzhny A. Ya. A method for identification of speech autoregressive parameters in frequency domain / A. Ya. Kalyuzhny, A. A. Kovtonyuk, V. Yu. Semenov // Acoustic bulletin. — 2010. — Vol. 13. — №2. — P. 20–27.
McLoughlin I. V. Line Spectral Pairs / I. V. McLoughlin // Signal Proces. —2008. —Vol. 88. — P. 448–467.