Digital audio/speech forensics

MPEG-1 Audio Layer 3 (MP3) is one of or if not, the most common audio media format these days and with advancement in technology, portable devices like MP3 players or digital voice recorders can easily record MP3 while on the go. Sometimes, these MP3 recordings might be important evidences that need to be presented in a court hearing. The challenge therefore is how to ascertain the audio file authenticity especially when MP3 files can be doctored easily with pervasive audio editing software. The purpose of this project is to implement certain techniques that can be used to detect forgeries, namely – deletion, insertion and substitution, in audio files. The project took reference from the paper by Rui Yang on detecting of forgeries in MP3 files via frame offsets. His algorithm based on LAME MP3 codec was tested on files doctored with the different forms of forgeries. After the testing, a smoothing technique based on the median filter concept was implemented to achieve better results of reducing the unwanted spikes and troughs present due to padding of zeroes by the codec, noise or silent frames. Next a novel method to determine the frame offset corresponding to the initial MP3 encoding of the file was introduced. Even though the technique was unable to detect any points of forgery, it will be able to detect different offsets from a file that has been encoded multiple times highlighting the fact it has been doctored before as non doctored files will only give rise to one single offset even after multiple encodings. As the method is brief, it might be useful to look further into detection of forgery in audios files that has been encoded multiple times.Bachelor of Engineering (Computer Engineering