High capacity real time broadcast monitoring

The authors address the problem of continuous real time monitoring of broadcast radio and television stations to identify when and where specific musical recordings and advertisements are transmitted. The technology has been deployed in a large scale network of remote monitors covering the United States. The method used involves a two stage pattern recognition procedure for eliminating unlikely candidate signatures with a minimum of computational effort. The first stage classification is biased to minimize the probability that a valid pattern will be missed while tolerating a relatively large number of false alarms. The second stage uses time warped matched filters to perform a confirming classification that minimizes false detections while maintaining a detection rate greater than 99%. The pattern recognition technique minimizes processor loading, allowing each remote monitor to search continuously for over 10000 candidate signatures