Melody Extraction from Polyphonic Music Combining Modified Euclidean Algorithm and Dynamic Programming

Melody extraction from polyphonic music is one basic and important task in the music information retrieval, and it is challenging due to the intrinsic complex nature of music. To more accurately describe the melodic salience and avoid the melodic contour shifting to wrong pitches within one note interval, the melody extraction method based on the modified Euclidean algorithm and dynamic programming is proposed in this paper. The modified Euclidean algorithm was used to estimate the frame-wise pitch candidates, and then the dynamic programming was introduced to model the salience and continuity constraints, and the final melodic pitch sequence was tracked. The performances of the proposed method were evaluated on three melody extraction evaluation datasets and compared with several other reference methods. The experimental results demonstrated that the proposed method achieved better melody extraction results, and outperformed the reference methods on all three datasets in terms of raw pitch accuracy.