The　auditory　quality　of　wideband　audio　is　generally　degraded　due　to　the　lack　of　the　high-frequency　in　network　transmission,　so　this　paper　presents　a　kind　of　audio　bandwidth　extension　method　from　wideband　to　super　wideband　based　on　local　least　square　support　vector　machine.　In　the　light　of　the　nonlinearity　of　audio　spectrum,　the　high-frequency　fine　spectrum　of　audio　signals　is　predicted　by　using　phase　space　reconstruction　and　local　least　square　support　vector　machine.　Combining　with　the　estimation　of　high-frequency　sub-band　energy　based　on　Gaussian　mixture　model,　the　proposed　method　can　effectively　recover　the　high-frequency　components　in　the　frequency　range　7　kHz~14　kHz　through　the　envelope　adjustment　of　high-frequency　spectrum　at　last.　Subjective　and　objective　testing　results　indicate　that　the　proposed　method　improves　the　auditory　quality　of　wideband　audio　and　outperforms　the　reference　methods　of　audio　bandwidth　extension.　©　2016,　Chinese　Institute　of　Electronics.　All　right　reserved.