Dictionary based inverted index for music information retrieval

Abstract

With the rapid progress in data storage and communication technology, there has been an explosive growth of music information. As the traditional metadata-based search engine can not provide natural and intuitive way to retrieve music, content-based music information retrieval has been attracting more and more attention and is becoming a popular research topic over the past decades. How to organize and search large scale music collections is one of the most fundamental challenging research problems. In this paper we introduce an inverted index approach well known in text information retrieval to organize and index music collections. According to musical form structure and repetition rule of musical themes, repeating patterns can be regarded as representation for musical themes. Statistical result indicates repeating patterns follow Zipf's law on a given music database. Music melodies are segmented into phrases based on musical dictionary composed of repeating patterns, and added into inverted index with necessary music information. Experimental results reveal that proposed index method outperforms the compared method greatly with much less storage space.