A　novel　watermarking　scheme　for　audio　authentication　is　proposed　in　this　paper.　The　scheme　combines　the　psychoacoustic　model　with　wavelet　packets　to　adaptively　select　the　best　basis　for　wavelet　packet　decomposition　such　that　the　subbands　are　closest　to　Critical　Bands,　given　the　temporal　resolution　and　computational　limit.　The　proposed　method　can　adaptively　control　the　watermarking　strength　in　the　wavelet　packet　domain　according　to　psychoacoustic　model,　which　ensures　both　transparency　and　sensitivity　of　the　watermarking　scheme.　The　amount　of　watermark　inserted　in　different　wavelet　packet　coefficient　is　based　on　current　subband　threshold　generated　by　psychoacoustic　model.　Generally,　in　our　scheme,　one　coefficient　carries　multi-bit　watermark　message,　which　corresponds　to　different　tampers.　The　new　scheme　can　authenticate　the　integrity　of　audio　signals　and　simultaneously　can　locate　the　tampered　region　in　temporal/frequency　domain　and　decide　the　tampered　extent,　as　well.　Experimental　results　show　that　the　proposed　scheme　has　the　desired　property　and　good　performance　for　audio　authentication.