An efficient approach to humming transcription for query-by-humming system

Query by humming is a straight approach to content-based music retrieval. There are three main challenges: query processing, melody representation and matching algorithm. This paper focuses on the first two problems by proposing a two-layer index and multi-stage matching to accelerate searching and improve effectiveness. The two-layer index includes tune features and profile features, which provides criteria for the coarse matching in the first two stages of matching, while in the last stage, we performs a precise matching. Experimental results demonstrate the feasibility of the proposed method.