Abstract: This paper presents a novel music fingerprint extraction method based on the harmonic characteristics of the beats in music. The harmonic information of each music frame is obtained, then based on the tracked beats, the mean of the harmonic’s information of all the frames in one beat is caculated, as the music fingerprint characteristic matrix of that beat. In order to improve the efficiency of music retrieval, a two level retrieval algorithm is proposed. The music whose beats in per minute（BPM） is close to that of the queried music clip is chosen as a candidate music, then the similarity between the music fingerprints of the queried music and of the candidate is calculated beat by beat, the music with the highest similarity is chosen as the retrieval result. Experimental results show that the music fingerprint and music retrieval algorithm proposed in this paper improves the retrieval accuracy and efficiency.

Key words: music fingerprint, beat, harmonic, music retrieval

摘要： 提出了一种基于节拍内音乐谐波特性的乐纹特征提取方法，首先求得每帧音乐的谐波信息，再利用跟踪得到的节拍，计算节拍内所有帧的谐波信息的均值，构成此节拍的乐纹特征矩阵。为了提高音乐检索的效率，设计了一个二级音乐检索算法：根据节拍信息，将与查询音乐片段的每分钟节拍数相近的音乐作为候选音乐，再逐节拍计算所查询音乐的乐纹和候选音乐乐纹的相似度，选择相似度最高的音乐作为检索结果。实验结果表明，提出的乐纹特征和音乐检索算法有效地提高了检索准确率和检索效率。

关键词: 乐纹, 节拍, 谐波, 音乐检索