An Enhanced STFT Segmentation Framework for ENF-Based Media Forensics

The electric network frequency (ENF) criterion has gained significant attention over the past two decades as a promising tool in digital media forensics. ENF is the frequency of the alternating current (AC) signal in a mains electricity network, exhibiting continual fluctuations within certain limits around a nominal frequency, contingent upon supplied and demanded power disparities. A sequence of ENF alterations is called an ENF signal, which is inherently embedded in audio and video recordings under certain circumstances. Several efforts have been made to accurately estimate the ENF signal from media. However, no matter how accurately estimated, a media ENF signal may not be reliably used in forensic applications unless sufficiently distinctive. To clarify, ENF may show similar fluctuation patterns at different time intervals. These patterns become more distinct over longer periods of time. Accordingly, working with as large an ENF signal as possible is critical for reliability. To achieve an extended and, thus, more distinctive ENF signal, this study proposes a smart segmentation scheme for Short-Time Fourier Transform (STFT)-based ENF estimation, which derives more data segments from a given media than the conventional STFT technique, leading to increased ENF estimates for any specified STFT parameter setting. The proposed approach can be combined with any ENF accuracy enhancement strategy to obtain relatively more reliable signals. Large-scale tests conducted with different STFT parameters and audio clip lengths showed that the proposed scheme can efficiently improve the performance when used alone or in conjunction with other ENF enhancement strategies.