Audio-forensic analysis of audio recordings through digital signal processing

The purpose of this research was to create an audio-forensics system that can be utilized to analyse audio recordings. Power lines have a notable effect on audio recordings, causing interference by injecting tones and harmonics at an approximate value of 60 Hz with deviations from this base frequency. These frequency variations can be used to identify certain information about the production of audio recordings through digital signal processing. Prior research shows that digital signal processing is a viable tool for timestamping an audio recording by comparing recorded audio signals with signals from the power grid. The goal for the first part of this research was to replicate the practice of time-stamping. For the second part of this research, the focus will was expanding beyond this replication to examine other information that can be attained through the analysis of signal frequencies and correlating power grid fluctuations. More specifically, the experiment will deal with methods of determining the location at which an audio recording as made. Currently, FM demodulation, which is used to attain the original information carried by a modulated signal, is viewed as a viable tool for this type of analysis because the power grid’s frequency is comprised of a carrier wave that experience modulation. This research experimentally verified these methods for time-stamping and geolocation. Measurements of ambient audio signals and power grid voltages were taken at various locations across campus, so the signals could be fingerprinted. Part of the process for location identification included a refinement of the algorithm for geolocation using the introduction of interharmonics; these interharmonics are frequencies that are part of power grid interference but are between harmonic frequencies. The last portion of my research include experimentation with fingerprinting my own audio recordings through the local injection of signals into the power grid. Overall, the research showed that characterization of signals through their frequency signature is possible, but future work is still required to refine the algorithm correlating audio and power grid signals.