Detection of audio interpolation based on singular value decomposition

Abstract

Interpolation attack, as a kind of tampering manipulation, is a common issue in digital audio forgeries. A new blind forensic approach for detecting interpolation forgery is proposed in this paper. As interpolation can lead to the statistical changes in the linear dependencies among digital audio sample points, singular value decomposition which can well express linear dependencies is used in the method. After audio is analyzed by singular value decomposition, a feature, counting the average number of zero singular values, is produced to describe the statistical changes. The proposed method based on singular value decomposition can locate the interpolation forgeries simply and fast compared to existing methods. The experiments conducted demonstrate the validity of the proposed method.