Similarity Matching Method for Music Melody Retrieval

In the research on content-based music retrieval, the fixed length of window is generally used to extract the features of melody. It has separated one fragment of music melody into multiple shorter window data. This will lose the whole feature of music melody and cause lower accuracy of retrieval. Therefore, current content-based music retrieval algorithms are discussed in detail and the emphasis is focused on the variable-length segmentation algorithm. The music retrieval algorithms for music fragments and humming fragments based on it are also analyzed. The former obtains the melody features of music fragments through accurate tonal values, calculated by the pitch extraction of single instrumental signals. The latter gets the pitch period of the humming fragments by calculation, which is based on the variable-length segmentation algorithm. Then the pitch period of each tone calculated by music retrieval algorithm and the results in experiments are used for similarity calculation. Compared to existing studies, the improved algorithm not only maintains the feature of low hit rate of variable-length segmentation algorithm, it also improves the matching speed and acquires better ability in fault-tolerance.