A Tampering Detection Framework for Digital Audio Signals Under Low-SNR Conditions

Extracting the electric network frequency (ENF) from digital audio signals is a crucial method in audio forensics. However, the ENF signal is highly susceptible to noise, making it difficult to establish an effective matching relationship with the reference frequency. This poses a significant challenge to the effectiveness of audio forensic frameworks. To address this problem, a low signal-to-noise ratio (SNR) digital audio tampering forensics (DATF) framework is proposed in this article. First, an improved chirp Z-transform (ICZT) method is proposed to extract the ENF signal from audio under low-SNR conditions. Subsequently, a dual-sampling isolation forest (DSIF) method is proposed to identify potential outliers by integrating bootstrap sampling with conventional random sampling. This approach enhances the perception of local data variations, thereby improving anomaly detection accuracy. Finally, a real-world digital audio dataset collected from low-SNR conditions is employed for tampering forensics. The experimental results demonstrate that the proposed DATF framework exhibits superior performance in ENF extraction, outlier detection, and tampering forensics compared with several state-of-the-art methods.