Query by Humming System Through Multiscale Music Entropy

摘要

Query by humming (QBH) is one of the most active areas of research under music information retrieval (MIR) domain. QBH employs meticulous approaches for matching hummed query to music excerpts existing within the music database. This paper proposes QBH system based on the estimation of multiscale music entropy (MME). The proposed technique exploits the statistical reliability through the MME for music signals approximation. Further, the Kd tree is employed for indexing MME feature vectors of music database leading to reduced search space and retrieval time. Later, MME feature vectors are extracted from humming query for recognition and retrieval of the corresponding song from music database. The experimental results demonstrate that the proposed MME and Kd tree-based QBH system provides higher discrimination capability than the existing contemporary techniques.