Audio tampering localization using modified iss watermarking in sparse-domain

This paper presents a blind audio tampering localization method in the perceptual sparse domain using a novel Modified Improved Spread Spectrum (MISS) watermarking approach. Current methods of audio authentication lack the ability to localize the tampered part of the audio in the presence of desynchronization attacks such as time shifting, scaling, etc. Our proposed method is able to localize the highly modified parts and then self-synchronize. To achieve this goal, Perceptual Matching Pursuit is used to compute a sparse and time-shift invariant representation of audio signals as well as 2-D masking thresholds. Then authentication code is inserted as a hidden watermark inside the sparse coefficient. To localize the tampered part of audio, the presence of watermark is detected inside the frames of input signal. To guarantee the high quality of watermarked audio, the watermark data is shaped by masking thresholds. To achieve high capacity, we propose a modified version of improved spread spectrum watermarking. In comparison to the conventional methods, experiments confirm the superiority of our approach in localizing the tampered parts of the audio and also confirm the high capacity of MISS signature insertion while preserving good audio quality.