关键词: 音频指纹,检索,熵,最大公共子串,编辑距离,动态时间规整

Abstract: A technology of identifying music using entropy based audio-fingerprint was introduced,which takes the music’s character of entropy as audio-fingerprint.In the domain of music identifying,the above audio-fingerprint enables to use flexible string matching algorithms.We adopted longest common subsequence (LCS),levenshtein distance and dynamic time warping (DTW) as the matching algorithms of this audio-fingerprint,and used a number of music as the test set.Every music has another performance which is generated from the original one,most of the other performances have been artificially changed,such as to be noise-accessed,accelerated,cut and so on,and some of them may even be paired of same music played by different orchestras.The obtained results are impressive,in which all the performances in the collection can be correctly identified either with LCS,levenhtein distance or the dynamic time warping (DTW) distances,proving the veracity,robustness and good distinguish ability.

Key words: Audio-fingerprint,Identifying,Entropy,LCS,Levenshtein distance,DTW