A Fast Audio Retrieval Method Based on Negativity Judgment

Abstract

A novel audio retrieval method based on negativity judgment is proposed. The foundation is that the similar audio feature vectors should be near after a transform procedure but which is a necessary condition but not sufficient condition to judge whether two audio segments are similar or not. So, it is reasonable to exclude audio vectors that can't meet the necessary condition for similarity judgment from retrieval matching, which can speed up retrieval effectively. In the paper, an audio feature vector quantization method called weighted self-similarity and an index method based on the idea of negativity judgment for Euclidean distance is introduced. Experimental results show that the method can search at a high speed.