A.Ya.Kalyuzhny, A.A.Kovtonyuk, V.Yu.Semenov
A method for identification of speech autoregressive parameters in frequency domain

Acoustic bulletin, Vol. 13 ¹ 2, (2010) p.20-27
The paper deals with considering of estimation of noisy speech autoregressive parameters in frequency domain. The method for calculating of likelihood function and implementation of the Expectation-Maximization method for the iterative estimation of speech autoregressive parameters have been illustrated on the base of frequency representation of the autoregressive signal. The analysis of different speech distortion measures shows that operation in frequency domain provides the similar accuracy as the corresponding approaches in time domain, but with essentially lower computational expenses. Moreover, a two-stage method for estimating of speech autoregressive parameters in frequency domain has been presented and investigated.
KEY WORDS:
speach signal, autoregressive parameters, the Fourier transformation, the Expectation-Maximization method, signal-to-noise ratio
TEXT LANGUAGE: Russian