In this paper a novel watermarking scheme for audio authentication is proposed. During quantization process, the watermark is embedded in the wavelet packet domain. The best basis of the wavelet packets adapts to the Psychoacoustic model so that the decomposed subband structure is closer to that of Critical Bands, which makes the embedding algorithm more flexible. Unlike previous schemes, the new method embeds several bits of watermark message into one coefficient. By introducing the detecting function, the proposed scheme can authenticate the integrity of the audio signals as well as locate the tamper in time/frequency domain. Effectiveness of the new approach is demonstrated by the experimental results.