°µÍø½ûÇø

High Dimensional Data and Music Genre Classification

Michael Brutz, Philip Lenzini

Applied Mathematics,Ìý

Date and time:Ìý

Thursday, July 25, 2013 - 3:45pm

´¡²ú²õ³Ù°ù²¹³¦³Ù:Ìý

Today there are more artists producing music than ever before. This is a trend that will likely always continue and the shear volume of pieces to pick from poses difficulties for anyone who wishes to search through this ever widening ocean of music. Creating a classification system for songs so that similar songs are grouped together is essential to this type of navigational process especially for anyone seeking a particular style of music. As music libraries grow, a classification method that does not require direct human interaction is not only useful but is becoming something of a necessity. In this talk, we will discuss a method for classifying songs by genre when given a set of testing data where we know the genres of the given songs. As the amount of data in a song's wav file can be very unwieldy, one of our foci is on a way to reduce the amount of data we have to handle to characterize a given song. Once the amount of data for each song has been stripped down, we then develop some way to get a notion of distance between different songs. Lastly, since the lines between genres are by no means well defined, we will discuss how to actually bin a song into a genre once the reduced level of data had been extracted.