A dual fragile watermarking scheme for speech authentication

In order to achieve the integrity authentication of speech files, a hash-based dual fragile watermarking algorithm is proposed. In the proposed scheme, a host speech signal is firstly transformed into a matrix, then each row and column of the matrix and a sensitive hash function (MD5) are used to generate two fragile watermarks. In the process of tamper detection, extracted watermarks and reconstructed watermarks are applied to construct two verification vectors, which form a verification grid that helps to detect attacks, locate the forgery domain and determine the type of modifications. Experimental results show that the watermarking scheme has good imperceptibility and fragility. More importantly, the proposed algorithm can not only detect attacks, but also accurately locate the tamped samples, and even identify the type of counterfeits.