Marsyas provides many feature extraction modules ("MarSystems"), including Mel frequency cepstrum coefficents (MFCCs). As this timbral descriptor is commonly used in many music information retrieval tasks, the first stage of this project will be to create a program using Marsyas to extract a single MFCC vector from each music file. The parameters of the extraction (sample rate, number of samples, number of coefficients, frequency range, etc) will be adjusted for optimal speed and accuracy of similarity detection.

June 12 update

Since the algorithm submitted by George Tzanetakis for MIREX 2007 was very successful, it is used for feature extraction in the first version of this program. The resulting vector contains 64 floats, describing a range of timbral features and their statistical distribution