|
As one of the important modules in query by humming system, humming feature extraction aims to extract the stable melody features. There are three typical personalization factors lead to instability of humming features, which are the base pitch shifting, the dynamic range changing and the humming rate changing. According to the characteristic of inconsistent humming rate, this paper mines the key point information that is stable to the change of melody rhythm. The melody is segmented and index is created first according to the key points, and the features in the segment are extracted. Then, a feature extraction method based on key point segmentation is proposed. Besides, using the theory of music theory and statistics, this paper takes the non-uniform segmentation statistic features and rhythm statistic features as the final humming features. Finally, numerical results are carried out to evaluate the validity and rapidity of the proposed schemes. |
|
Keywords:query by humming; key point; feature extraction; melody feature |
|