Audio fingerprint is a digital digest of the signal which represents the most significant characteristics and is widely used in identification、security validation and integrity verification of audio signal. This paper presents a method of generating audio fingerprint based on the power spectrum characteristics of the signal. After calculating STFT of the input sample and extracting the peak of the power spectrum, we use the space structure of the peaks as input of the hash function of the audio fingerprint algorithm. As experiment shows, it could achieve a recognition accuracy of over 95% when the audio sample lasts more than 2 s and shows good resistance to AWGN noise. It is easy to implement、robust enough and exhibits fast and accurate identification ability.