Match Recognition System 1.2 2006/12/19 14:01:07 US/Central 2007/01/08 11:22:24.972 US/Central Andre T. Mosley dremos@rice.edu Yu-Heng Jaret Lee jaret.lee@rice.edu John Broadway jtb5020@rice.edu Po T Wang ptwang@rice.edu Andre T. Mosley dremos@rice.edu Yu-Heng Jaret Lee jaret.lee@rice.edu John Broadway jtb5020@rice.edu Po T Wang ptwang@rice.edu Richard G. Baraniuk richb@rice.edu Mark A. Davenport md@rice.edu Match Recognition System This module explains how we implemented the matched filter phase of our audio recognition system. To implement the Match Recognition System for our program, we used a basic matched filter that compares the input song’s fingerprint to each fingerprint in the database. Because we are using compact representations for each song, the matched filter will not take an insane amount of time to complete. In order to improve the speed of the algorithm even more, we decided to implement the filter in the frequency domain since it is involves simple matrix multiplication. First of all, the input song goes through the Audio Fingerprint Generator so that we have its compact representation. Next, our system normalizes this representation, takes its FFT, and compares it to the normalized FFT of each of the fingerprints in the database. At this point the fingerprints are all in the frequency domain, so the only thing left to do to implement the matched filter is to multiply the input song’s representation with each one of the database representations. This process requires zero-padding the shorter signal in the comparison so that the two matrices are the same dimensions before actual multiplication. The max(max()) function in Matlab takes the maximum value in the matrix, which corresponds to the matched filter “spike”. The database signal that returns the largest spike when compared to the input signal is our most probable match and the database index of this matched song is used to index the name of that corresponding song from the array that holds the names of all database songs. Our system also does this process for the second highest match. The output of the Match Recognition System is just the two most probable match titles, and their corresponding spike values. This allows users to see the confidence level for each match.