Dimension-Reduction Technique for MPEG-7 Audio Descriptors

The MPEG-7 audio signature descriptors can be used to identify whether one piece of sound track is the same as another one in the database. This technique, known as music identification, can be applied to broadcast monitoring and copyright control. However, a practical database usually has a lot of sound tracks, therefore it requires too much time to directly compare the descriptors in the database with the unknown ones. In this paper, we propose an averaging method to reduce the dimensionality of the descriptors to a much lower degree. Our experiments show that the dimension-reduced descriptors still have a high discrimination capability. Using these descriptors, an efficient dual-resolution search method can be implemented for music identification.