An　LP　(Linear　Prediction)　spectrum　modification　method　is　proposed　based　on　linear　extrapolation　for　different　signal　to　noise　ratio　under　the　white　noise　environment.　A　direct　modification　on　the　noisy　LP　spectrum　is　employed,　since　the　autocorrelation　coefficients　of　the　original　speech　are　not　available　in　some　practical　applications.　The　linear　suppression　rule　is　applied　into　the　high-frequency　region　of　the　noisy　LP　spectrum,　while　no　modification　is　made　in　the　low-frequency　region.　The　experimental　results　on　ITU-T　G.722.2　demonstrate　that,　comparing　with　the　reference　methods,　the　proposed　method　could　provide　the　reconstructed　speech　with　an　obvious　improvement　both　in　objective　and　subjective　quality　by　using　the　modified　LP　coefficients.