A　nonlinear　audio　bandwidth　extension　method　based　on　hidden　Markov　model　(HMM)　is　proposed　to　reconstruct　super　wideband　audio　signals　from　wideband　audio　signals.　The　sub-band　energy　of　high　frequencies　is　estimated　based　on　HMM　according　to　the　low-frequency　features　of　audio　signals.　The　smoothness　of　energy　transition　for　the　extended　audio　signals　can　be　improved　in　time　domain　and　frequency　domain.　In　addition,　the　fine　information　of　high-frequency　components　is　recovered　to　guarantee　the　timbre　of　the　extended　audio　by　nonlinear　prediction　based　on　the　nearest-neighbor　matching.　The　objective　and　subjective　test　results　indicate　that　the　proposed　method　outperforms　the　conventional　methods　including　blind　bandwidth　extension　and　nonlinear　extensions.