In this paper, we present three methods for improving the searching speed and accuracy for query by humming (QBH) system with large melody database. 1) At the feature level, to minimize the inevitable errors caused by a single pitch extractor, three different pitch extraction algorithms are fused together to gain more credible and robust pitch sequence. 2) To speed up the matching process, a candidate set reduction method is firstly adopted to filter out the unlikely candidates by faster but less precise methods; then a more accurate but slower strategy is executed on the survival candidate set to perform a finer match. 3) At the decision level, we utilize these scores generated during the filtering stage and fine-matching stage to fusing together to get more accurate result. The proposed system achieved mean reciprocal rank of 0.929 for the corpus used in MIREX2006  while cost an average of 0.58 seconds for one query. The results reveal the advantage of our system on speed and accuracy comparing to other system participated in that contest.
Bibliographic reference. Wang, Lei / Huang, Shen / Hu, Sheng / Liang, Jiaen / Xu, Bo (2008): "Improving searching speed and accuracy of query by humming system based on three methods: feature fusion, candidates set reduction and multiple similarity measurement rescoring", In INTERSPEECH-2008, 2024-2027.