Results and Conclusions 1.2 2006/12/20 17:31:36 US/Central 2006/12/22 09:48:00.165 US/Central John Broadway jtb5020@rice.edu Yu-Heng Jaret Lee jaret.lee@rice.edu Andre T. Mosley dremos@rice.edu Po T Wang ptwang@rice.edu John Broadway jtb5020@rice.edu Yu-Heng Jaret Lee jaret.lee@rice.edu Andre T. Mosley dremos@rice.edu Po T Wang ptwang@rice.edu Richard G. Baraniuk richb@rice.edu Mark A. Davenport md@rice.edu This module sums up the results of our audio file recognition system. Using a 46-song database, we found our system to be quite reliable, even with short clip lengths and low signal-to-noise ratios. The system was able to detect matches even when the identity of the song clip could not be audibly ascertained. Since the system handles distortion well, it should be well suited to receive input clips from a microphone.

SNR vs. Confidence Level for Match. Confidence is the ratio of the calculated matched filter spike to the ideal spike. Even with the lowest SNR we tested (200% noise), matching still occurred correctly.

Clip Length vs. CPU Time. As clip length increases, match-seeking time increases linearly. Smaller clips are faster. 75 seconds minimum time is required to search through the database. Even with the shortest clip tested (1 second), matching occurred correctly. The main drawbacks of our system were in its computation time. A little over a minute is required to perform the search and matching operations. With many more songs added to the database, the delay would be considerably longer.

Computation Time Breakdown. The FFT algorithms in the Match Recognition System consume most of the operation time. If the database were large enough to use the system in practical applications, the search delay would be far too long. To improve the operation time of the system, the Match Recognition System would most likely be optimized. Possible methods for this optimization include predictive shifting in the filtering stage and partitioning of the database.