Music main melody extraction by an interval pattern recognition algorithm

Main melody in one piece of music serves as a significant feature for music style analysis and music information retrieval. This paper proposes a dictionary-based algorithm to extract the main melody from numerical music scores. Classical music is chosen as our target of research. First, the melody line that can best represent the original music is extracted from the sophisticated score which contains various harmonies and instruments. Then, inspired by implication-realization theory, the direction of intervals instead of notes are used as input of our proposed algorithm, which is adapted from the classic algorithm Lempel Ziv 78 (LZ-78), to generate a dictionary that has numerous repeated patterns. After that, some post-processes including elimination, splicing, and extension are implemented on the dictionary to obtain one or more main melodies.