Digital audio forensics

Abstract

In the research field of Digital Audio Forensics, many researches have been done in Speech Recognition, Speaker Recognition, Audio Enhancement, but only a few can be found in Speaker Environment Recognition. This is due to the diversity of the environment, and the complexity to analyse it.

This project investigates a method to analyse the speaker environment using MFCC and selected MPEG-7 audio descriptors. The author also studied and compared the results between several supervised and unsupervised classifiers like GMM, k-NN, Naïve Bayes, etc. used in the model to determine the best classifier. Other feature like Zero Crossing Rate was also investigated in the effort to improve the performance of the model.