Music humming retrieval method based on melody feature clustering optimization
The existing information retrieval methods are difficult to accurately capture and match the melodies hummed by users.Therefore,a humming retrieval method is proposed,which uses the digital interface audio of music instruments to construct a melody feature database,extracts pitch vectors through frame sampling technology,and stores them in combination with metadata.The study adopts optimized clustering algorithms to improve feature discrimination and further deepens feature extraction using neural networks.During the retrieval process,combining the least squares method and dynamic time warping algorithm to quickly and accurately match the melody hummed by users.The research results indicate that the optimized K-means algorithm performs well in humming retrieval,with accuracy and recall rates of 0.79 and 0.89,respectively,and high F1 values.On the intra class sum of squares metric,its score is below 3.000,surpassing other algorithms,indicating that the research has improved the efficiency and accuracy of music humming retrieval,providing a new practical tool for music information retrieval technology.
万方数据
维普
