Tampering detection in digital audio recording based on statistical reverberation features

Abstract

Audio authentication has become a challenging task with an increase in easily available manipulation tools which can be used to forge the audio recordings. This paper focuses on audio tampering detection which is a task that involved in the area of audio forensics. Reverberation-based acoustic features are considered to distinguish between original recordings and their tampered versions. Two types of feature sets have been considered which depend on the decay rate distribution of the signal in each frequency band. The statistical features of the decay distribution and that of the Mel-frequency cepstral coefficient matrix of the reverberant component have been used. A threshold-based technique that considers the percentage error between these statistical parameters of the original audio recordings and their tampered versions has been employed. The methodology has been tested on synthetically created data set which consisting of original recordings and their tampered versions recorded in different acoustic environments.