Detecting Audio Splicing Forgery Algorithm Based on Local Noise Level Estimation

摘要

Splicing including homogenous splicing and heterogeneous splicing is one of the most common ways to tampering digital audio. This paper focuses on the study of heterogeneous splicing. Since different audio is recorded at different background environments, there exist significant differences in background noise level for different recorded audio. In this paper, the level of background noise is used to detect the splicing tampering in audio. However, unlike other methods that use background noise level for tamper detection, we do not detect background noise level after continuous framing. In order to more accurately determine the location of tampering, we first need to determine the length of each syllable in the audio using endpoint detection method. Here, we use the spectral entropy method to determine the length of each syllable. Then the variance of background noise of each syllable is calculated according to the kurtosis value characterized the number of peaks in the density distribution curve at the mean that is close to a constant in the band-pass filtered domain. Finally, by comparing the similarities between the variance of background noise of each syllable, it is judged whether there is an operation of heterogeneous splicing tampering in the audio. Our algorithm can accurately locate the heterogeneous splicing syllables.