The Weighted Sum of the Line Spectrum Pair for Noisy Speech

Linear prediction (LP) provides a robust, reliable and accurate method for estimating the parameters that characterize the linear time-varying system. It has become the predominant technique for estimating the basic speech parameters such as formants. Linear prediction exploits the redundancies of a speech signal by modelling the speech signal as an all-pole filter. The filter coefficients are obtained from standard autocorrelation method or covariance method. In this work, we focus on a new developed method, the Weighted Sum of the Line Spectrum Pair (WLSP), which is based on the Line Spectrum Pair (LSP) decomposition. In conventional LP, the LSP decomposition is computed to quantise the LP information. WLSP utilises the LSP decomposition as a computational tool, and it yields a stable all-pole filter to model the speech spectrum. In contrast to the conventional autocorrelation method of LP, WLSP takes advantage of the autocorrelation of the input signal also beyond the time index determined by the prediction order in order to obtain a more accurate all-pole model for the speech spectrum. With the help of the spectral distortion method, the experiments results show that WLSP can distinguish the most important formants more accurately than the conventional LP.