Method of Increasing the Audio Fragment Search Speed using kd-trees

Abstract

The article is devoted to solving the problem of audio identification by the short fragment. The method of the nearest template search for audio fragment (5s) based on kd-tree was proposed in the paper, which allowed to reduce the computational complexity significantly compared to full search. To perform a quick search based on kd-tree in pre-formed body of templates algorithms and software were developed. Analytical relation for evaluating the proximity between the unknown fragment and templates was proposed, which is based on calculating the distance from the fragment to the specified number of the nearest centroids.