Music in the Real World: Live Music Retrieval and the Limitations Thereof

Music is everywhere around us. Cliché, yes, but music has found a way to infiltrate nearly every field of study known to man. Writers and authors include musical references to give context to novels and works of literature. Physicists study frequencies of music and sound to determine their effects on animals and humans. The field of Information Science is no different when it comes to its relationship with music. Music Information Retrieval (MIR), a subset of Information and Library Science, was founded in 2000 to gain a better understanding of the information music contains and how this can best be extracted for human use (Byrd, Fingerhunt, 2002). MIR deals with both the metadata of music (band, lyrics, album name, etc.) and actual content of music (chord structure, melodic design, and rhythmic analysis). As the title suggests, a large part of MIR deals with retrieval, which is done by both humans and machines. Examples of retrieval include the applications of SoundHound and Shazam, which make use of audio fingerprinting technology in order to query databases that contain songs in order for a user to identify a song quickly and efficiently. As stated above, the field of MIR is relatively new with respect to academia, with less than 15 years worth of research and information available. Although the term itself was originally coined in a lecture in the late 1960s, the field itself did not receive formal recognition until a much later time. Because of this, many areas of MIR have not been studied to their fullest extent. One of these areas, content-based music retrieval in a live environment, has seen painfully little research. It is this area I wish to study more in-depth by posing the following questions: How successful are current content-based music identification systems at identifying jazz songs performed live? In cases where they fail, what are the possible causes of those failures? What can be done to improve the effectiveness of these applications and methods and what does this imply for future research? As I stated before, the lack of research done for live music applications leads me to believe that more should be done. The need for identification of tracks occurs more often in live venues, like concerts, more so than studio recordings How many times have you heard from friends or neighbors, “What song is this? Do you know this one?” By looking at identification methods and their limitations, research can address these issues and hopefully design more efficient applications.Bachelor of Scienc