Aiming at the problem that the audio steganography has low utilization of carriers, poor imperceptibility and small embedding capacity, an adaptive audio steganography based on the wavelet packet decomposition and matrix code was proposed. Comparing the differences between the wavelet-packet decomposed coefficients before and after the audio's MP3 compression, the algorithm took the unchanged bits' position of wavelet-packet decomposed coefficients as embedding carriers, which effectively increased the embedding capability. And the algorithm improved the matrix code by using chaotic model to generate random triple-groups, which promoted the safety and efficiency. As for capacity, the proposed algorithm promoted about 30%, compared with the algorithm that directly uses the medium-frequency sub-bands as the carriers. On the aspect of Signal-to-Noise Ratio (SNR), the proposed algorithm promoted about 9%, compared with the matrix steganography that fixes the triple-groups. The experimental results show that the algorithm is correct and can basically satisfies large-capacity and secure communications.