Two-layer Distance Scheme in Matching Engine for Query by Humming System

Feng Zhang, Yan Song, Lirong Dai, Ren-Hua Wang

Abstract: In query-by-humming system, minimizing the impact introduced by the pitch tracking errors and humming errors is always a difficult problem. In this paper, we propose a two-layer distance scheme instead of the global DTW distance in traditional QBH system, which is motivated by the people’s perception of humming. The local distance measure tries to find the correct part of humming, which is robust to the hummed error and the noise. Also, the approach to combine the local and global distance measure is proposed. The experiment of our QBH system on the mobile phone channel shows that the performance is greatly improved.