Melody Matching Model Based on a Genetic Algorithm

秦静,王醒策,周明全,刘新宇
DOI: https://doi.org/10.3969/j.issn.1006-7043.2010.02.020
2010-01-01
Abstract:Content-based music information retrieval(MIR) has received increasing attention in the field of signal processing and pattern recognition.Methods for feature selection,representation and matching are three key points in the MIR process.Some physical characteristics of musical features are favored in the perceptions of listeners.As a result,melody is thought to be the most important feature.Pitch tracking and dynamic threshold note segmentation were used to formulate a template for queries about music.A melody contour alignment algorithm based on a genetic algorithm(GA) was proposed,which was used to align and correct the input pitch template.A weighted combination of the Euclidean measure and the dynamic time warping(DTW) algorithm was used to make the search process fuzzy and more effective.The experiment demonstrated that the algorithm's tacking speed and accuracy could meet the needs of a query-by-humming system,and could be used in similar systems.
What problem does this paper attempt to address?