Similarity matching of continuous melody contours for humming querying of melody databases

Music query-by-humming is a challenging problem since the humming query inevitably contains much variation and inaccuracy. Many of the previous methods, which adopt note segmentation and string matching with dynamic programming, suffer drastically from the errors in the note segmentation which affects retrieval accuracy and efficiency. In this paper, we present a novel melody similarity matching technique which is based on continuous melody contour. The continuous melody contour has many advantages over previous string based melody representation. We introduce a contour alignment technique which addresses the robustness and efficiency issues. We also present a new melody similarity metric which is performed directly on continuous melody contours of the query data. This approach cleanly separates the alignment and similarity measurement in the retrieval process. We show how to robustly and efficiently align the query melody with the reference melodies and how to measure the similarity subsequently. Our melody alignment method can reduce the matching candidate to 1.7% with 90% correct alignment rate. The overall retrieval system achieved 80% correct retrieval in the top 10 rank list.