A Study of Musical Features for Melody Databases

Music has an auditory and temporal nature. The same piece can be interpreted in multiple, and often unrelated ways. Together with the limitations of its representations, the design of content-based music retrieval systems on the Web is a challenge. Most literatures on the problem map the problem to existing information retrieval paradigms, mainly that of text, by modelling music as a sequence of features Lin77]]MS90]. However, this mapping raises issues to be solved. Through the study of the statistical properties of six features including Proole, Note Duration Ratio Sequence, Interval Sequence and their variants, we answer four important questions that arises from mapping. These include the number of musical \alphabets" and \words", whether Zipf's law holds for musical features, whether there are any musical \stopwords", and the range of n for n-gram based music indices. The answers to these questions will aaect crucial parameters in music retrieval systems, and whether the mapping and techniques used in text retrieval is eeective for music. They will also aaect the way to reduce the sizes of musical indices, which we will show is important for music retrieval systems.