Current　statistical　audio　watermarking　can＇t　guarantee　imperceptibility.　The　proposed　scheme　in　this　paper　uses　a　statistical　method　to　embed　watermark　in　the　tonal　components　in　wavelet　packet　domain　and　integrates　psychoacoustic　analysis　to　control　watermark　strength,　which　makes　watermark　inaudible.　Error　detection　rate　analysis　and　experimental　results　show　that　the　proposed　method　is　inaudible　and　sustainable　against　common　signal　processing　manipulations　and　malicious　attacks.