哼唱检索系统中一种有效的旋律匹配方法

An Effective Method of Matching Melody in QBH System

针对音乐数据库中有些歌曲不符合音阶模型算法的问题,提出了一种将音阶模型与均值法相结合的改进音调调整算法,对不符合模型的哼唱输入和midi使用均值法进行音调调整,解决了大数据库情况下的音调调整问题。同时针对人的哼唱和pitch提取等处理可能存在错误的问题,提出了一种新颖的两层DTW距离度量方案,大大提高了最终匹配的准确度和系统的鲁棒性。实验结果也表明了改进的音调调整算法和两层DTW距离度量方案的有效性。

Some songs in the music database do not accord with the scale model. To solve this problem, the paper proposes an improved key transposition algorithm by combining scale model algorithm with average method. Query input or midi which doesn＇t fit model can be processed by the average model. The new algorithm can be used in larger database. The paper also proposes a novel two - layer distance scheme for the pitch tracking errors and humming errors, enhancing the accuracy of the final match and the robustness of the system greatly. The experiments of improved key transposition algorithm and two- layer distance scheme show these approaches are effective.