Accuracy enhancement in environment sound recognition using ZC features and MPEG-7 with modified K-NN classifier feature

摘要

In this paper, we modify the K-NN classifier feature for environment recognition from audio particularly for forensic application. We compute the distance between the first frame from the testing file with all frames from the training file, instead of only the corresponding frames, then we take the average. We investigated the effect of temporal zero crossing feature and some selected MPEG-7 audio low level descriptors on environment sound recognition. Experimental results show that higher recognition accuracy is achieved by using the modified K-NN classifier and confirm that the accuracy is increased when the size of the training file is decreased.