Sixth International Conference on Spoken Language Processing
Feature extraction of speech has been an important problem on speech processing. We are deeply interested in time-varying complex speech analysis method and we have already developed two kinds of time-varying complex AR (TV-CAR) parameter estimation algorithms for analytic speech signal, which are based on minimizing mean square error (MMSE) and Huber's robust M-estimation. This paper presents new robust parameter estimation algorithm for the TV~CAR model on the basis of an instrumental variable (IV) approach that may be robust against additive noise. The experiments with natural speech corrupted by white Gaussian demonstrate that the time-varying complex AR method achieves robust spectral estimation against additive white Gaussian owing to the improved resolution in the low frequencies.
Bibliographic reference. Funaki, Keiichi (2000): "A time-varying complex speech analysis based on IV method", In ICSLP-2000, vol.4, 688-691.