Music Information Retrieval System Using Lyrics and Melody Information
Tao Wang, Dong‐Ju Kim, Kwang‐Seok Hong, Jeh-Seon Youn · 2009
Multimedia content can be described in versatile ways as its essence is not limited to one side. For music data these multiple fields could be a songpsilas audio features as well as its lyrics. But most recent research revolves around melody information for retrieval. Therefore, we proposed an MIR system that utilizes the userpsilas acoustic signal from a singing voice and retrieves the music information using both lyrics and melody information. The lyrics recognition module uses a keyword spotting system based on text-content of the lyrics by an HMM comparison engine. The melody recognition module extracts pitch and MFCC features from the user singing input and then retrieves music by a GMM comparison engine. Consequently, the proposed MIR system consists of fusing the lyrics and melody recognition module in which the melody recognition especially operates to restrict recognition candidates. Experiments show that the proposed MIR system has recognition rate of 72.72% to 83.64% when the numbers of restricted recognition candidates are from 10 to 50.