A construction method of music features libraries in QBH

How to construct music feature libraries properly is significant to Query by Humming (QBH) in a cross-media information retrieval (CMIR) system. On the foundation of content-based retrieval for audio media such as speech, music or song, sound, this paper divides the music features metadata into three parts: MIDI pure structured (PS) music data, WAV real example (RE) music data and keyword sound (WS) text-information data. Firstly, we introduce our system architecture. Secondly, we propose anew construction method for music feature libraries in order to achieve more satisfactory results. Thirdly, we present and design the novel mix-indexing architectural structure in order to carry out some efficacious searching algorithms. At last some pilot experiment results show the feasibility of our study work.