VICTOR: An Adaptive Framing-based Speech Content Authentication and Recovery Algorithm

Speech authentication plays an important role to verify the integrity of speech in multimedia security. Nowadays, the existing speech authentication methods segment all speech content to fix-length sub-frames. Then embed watermarks in every sub-frame. However, not all speech content needs to authenticate its integrity such as silent speech. Therefore, the fix-length framing methods will cause needless computation and even affect the inaudibility. To resolve these drawbacks, VICTOR: a novel adaptiVe framIng-based speeCh conTent authenticatiOn and Recovery algorithm is proposed in this paper. VICTOR first obtains the non-silence speech frames from the original speech signals using a pitch tracking method. Then compressed data of these non-silence speech frames are computed by the designed hybrid compression algorithm. After that, the watermarks are generated by connecting the compressed data and frame ID, which are embedded into the LSB of another sub-frame. In the authentication process, the tampered speech frames can be located and recovered using the embedded watermark by self-synchronization. Analysis and experimental results demonstrate that VICTOR can detect various attacks and it can effectively also recover the tampered content.