Abstract: As an important way of music retrieval， query by humming has gained wide attention because of its effectiveness and convenience. A novel retrieval technology of humming was proposed based on sentence features， which could provide fast retrieval for query by humming. In the proposed technology， the music database and humming given by users were first partitioned according to natural pauses， and then the BDTW (bounded dynamic time warping) algorithm was adopted to compute pitch similarity. In addition， users can also establish personalized settings in accordance with their own humming， and limit the maximum local length variance between music database fragments and query sequences. In addition， the index structure DIS was established to support music sentence query， which could reduce searching time. The experimental results verified both the efficiency and effectiveness of the proposed retrieval method.

Key words: music retrieval, query by humming, whole matching, subsequence matching, dynamic time warping (DTW) algorithm