Comparison Of Features For DP-Matching Based Query-by-Humming System

In this paper, we compared three kinds of similarity measures for DP-matching based query-by-humming music retrieval experiments. First, a DP matching-based algorithm is formulated using the similarity between a deltaPitch of an input humming and that of a song in the database. Then the three similarities are introduced: distance-based similarity, quantization-based similarity and fuzzy quantization-based similarity. The three similarities are compared by experiments. From the experimental results, the distance-based one gave the best recall rate. In addition, we examined the combination of distance-based and fuzzy-quantization-based similarities. The experimental result showed that the recall rate was improved by the combination.