哼唱检索系统中的特征提取研究

Feature extraction module has a very important position in the query by humming system, whether feature selection is highly independent, distinguished and representative do great influence on the system retrieve performance. This paper provides details on some of the feature extraction algorithm, based on which a pitch and note feature extraction is described, it adopts the artificial tagging methods and DTW, EMD matching algorithm to make evaluation of feature selection, get the precison of the fundamental frequency extraction and note segmentation, and finally from the perspective of the overall evaluation of retrieval systems to get the system retrieval accuracy property.