Majed El Helou, Abdel Wahab Turkmani, Rawan Chanouha and Samer Charbaji, American University of Beirut, Lebanon

The electric network frequency (ENF) of power lines leaves its trace in nearby media recordings. The ENF signals vary in a consistent way in a given power grid. Therefore, it is possible to develop signal processing and machine learning techniques to identify the grid of origin by extracting attributes of the embedded ENF signal in recorded audio. This paper presents a model based on a novel ENF extraction technique with training on audio and power recordings from different grids. The proposed approach is based on correcting erroneously selected peaks from the Short Time Fourier Transform (STFT) by leveraging time correlations. These peaks are mistakenly taken for the frequency component belonging to the embedded ENF signal of the power grid and are corrected by the algorithm. Results on a test set of 50 recordings from nine different locations demonstrate the effectiveness of the proposed approach with an overall accuracy of 88%.

ENF, Region of Recording, Media Recordings, Power Grids, Noise in Audio Recordings.