Traditional　threshold　de-noising　algorithm　in　wavelet　domain　is　effective　only　for　reducing　white　noise.　In　order　to　eliminate　many　kinds　of　noises,　a　speech　enhancement　approach　with　adaptive　threshold　based　on　bi-orthogonal　wavelet　packet　decomposition　is　proposed　in　this　paper.　This　method　updates　the　threshold　by　adaptively　tracking　the　level　of　the　noise　and　a　dynamic　threshold　is　used　for　reducing　noise.　Experimental　results　show　that　the　proposed　approach　outperforms　orthogonal　wavelet　based　de-noising　approaches　for　speech　enhancement　and　can　efficiently　suppress　three　kinds　of　noises.