Linear predictive analysis of noisy speech

New methods are proposed to improve the robustness of the linear predictive (LPC) analysis of speech to noise. These methods are developed based on the concept of noise compensation. By subtracting the noise power from the autocorrelation function of speech iteratively, the proposed methods achieve success in both reducing the bias of the LPC coefficients and guaranteeing the stability of the LPC inverse filter.