Abstract: This paper presents a digital signature algorithm based on feature of the audio.Extract the statistical feature of wavelet domain of the chaotic encrypted initial audio as the eigenvectors，which is then Hashed to generate digital signature.The zero-watermark algorithm is used to embed the digital signature to the original audio to solve the save，the management and the transport of the digital signature.When authenticated，it is used related detection technology to get the quantitative evaluation of the similarity coefficient of the signature.The experiments confirm that the signature in this algorithm has robustness to the activities of preserving the audio’s content，and is sensitive to the changing of the audio，and it realizes the authentication of the audio.