Study On Identity Verification And Content Authentication Of Speech Based On Perceptual Hashing

With the rapid development of the multimedia technology and network technology, the distribution and storage of digital multimedia information become more and more convenient. However, these conveniences also bring some disadvantages such as illegal copy and malicious tamper. Therefore, multimedia protection and authentication become more and more important. As a hotspot of multimedia information security, perceptual hashing shows perceptual property and robustness to common signal processing, so it can be applied in the speech authentication.This dissertation mainly focuses on identity verification and content authentication for speech based on perceptual hashing. The main work is given as follows:(1) A speech identity verification algorithm based on the perceptual fingerprinting hashing is proposed. First of all, a chaotic key is used to select some rectangles of the fingerprint, and then the gravity of each rectangle should be calculated. After quantified, the gravity will be converted to binary. The perceptual hashing is the sequence of these binaries. In order to associate the speech with the perceptual hashing which stands for the feature of the speaker, as one kind of digital watermarking, the perceptual hashing will be embedded into the speech.Therefore, the algorithm overcomes the lack of the traditional identity verification schemes which are easy affected by the environment and the human body status. The algorithm is robust about noise, rotation and so on. And it has strong security and uniqueness.(2) Based on the perceptual hashing of the speech ZCR (zero-crossing rate) and speech centroid, a speech content authentication algorithm is proposed. Through perceptual hashing extract function, the perceptual hashing which stands for the feature of the speech can be generated. Through the method of mean quantization, as seem as one kind of digital watermarking, the perceptual hashing will be embedded into the speech. When authentication, the perceptual hash should be extracted from the speech by the extraction algorithm, firstly. And the perceptual hashing of the speech should be calculated. Finally, whether the speech has been tamper and which area has been tamper can be determined by the authentication algorithm.The experiment results show that the algorithm can resist the common processing, for example the noise and low-pass filter, and it can be able to determine and locate the malicious tampering attacks correctly.