N. Patel and P. Mundur (USA)
music, information retrieval, humming and pattern finding
In this paper, we present a methodology to detect the major repeating patterns of a musical score using a novel approach based on techniques from the traditional text based Information Retrieval systems. We propose an n gram based approach to determine the repeating patterns in a musical composition. Finding major themes in music is critical for any large Music Information Retrieval systems that have to be scalable. Queries performed against this subset of the music database as against the entire corpus lead to a great improvement in scalability. Research in the area of detecting themes from music scores has been mostly based on pattern matching techniques. The current research performed is mostly limited to analysis of music which is taken from a MIDI file. This excludes music not available in MIDI format which means excluding a large number of musical pieces from the database. Research is being done in converting sound files into the MIDI format, but, as of now, the science has not progressed to such a state that we can get a perfect translation from any other format to a MIDI format. We perform our work on standard WAV files which contain PCM encoded data allowing any piece of music to be put into the system.
Important Links:
Go Back
