Environment Sound Recognition for Digital Audio Forensics Using Linear Predictive Coding Features

Linear Predictive Coding coefficients are of the main extraction feature in digital forensic. In this paper. we perform several experiments focusing oil the problems of environments recognition from audio particularly for forensic: application. We investigated the effect of temporal Linear Predictive Coding coefficient as feature extraction on environment sound recognition to compute the Linear Predictive Coding coefficient for each frame for all files. The per.. formance is evaluated against varying number of training sounds and samples per training file and compare with Zero Crossing feature and Moving Picture Experts Group-7 low level description feature. We use K-Nearest Neighbors as classifier feature to detect which the environment for any audio testing file. Experimental results show that higher recognition accuracy is achieved by increasing the number of training tiles and by decreasing the number of samples per training file.