Content-based Music Information Retrieval using Pitch Histogram of Band Pass Filter Signal

Recently, content-based techniques using a signal itself as a query have been required to efficiently identify an unknown music signal. This paper addresses a solution for “query by example” such as the content-based music information retrieval. In this work, only a few seconds’ duration of the music signal is used to recognize the unknown polyphonic music by matching it to preindexed audio references. Here, to consider a real-world situation, the query signal is recorded through a general microphone when music sound is played by MP3 player. Thus, we propose a robust and efficient identity matching method using melody pattern so that the distortion in frequency domain is reduced. The experiment shows that the proposed technique is generally effective in the case of retrieval for a query signal recorded in real-world condition. Also, it can be successfully applied to music identification with only a few seconds’ duration of the unknown music signal.