Research On Key Technologies Of Speech Perceptual Authentication

With the rapid development of communication techniques,speech has played an increasingly important role in remote comlnunication.At the same time,the openness of wireless network channel attracts attention of malicious attacks,such as eavesdropping and tampering.That makes the application of authentication system urgent.But the authentication of speech is different from that of text,due to the large data with high redundancy of speech information.Otherwise,there are strict efficiency requirement foro authentication systems,for the resource of communication terminals is limited and processing of speech is real-time.At present, the speech authentication algorithms are rare.This paper focused on the research of speech authentication system based on perceptual characteristics,and all aspects of that,from model,algorithms to realization have been discussed.The main work and innovations are as follows:1.A model of speech perceptual authentication was proposed to meet with current requirements.Based on the analysis of all aspects of performance and application modes of perceptual signature,a general model was constructed,which combined the speech perceptual digest with the public key signature algorithm,to achieve the authentication of both content and entity.A framework of algorithms was presented. Principles and methods of each part were discussed for following design.2.A speech perceptual signature combined with masking effects were proposed. The part of feature extraction was formed with MFCC and masking effects.MFCC, which is a wide used feature extraction algorithm,could not be applied in the domain of content authentication of speech for the deficiency of robustness.In this paper,the elimination of redundancy has further enhanced the robustness of MFCC features. Then,SVD algorithm was used to construct the digest,and AES and Rainbow algorithms were used to encrypt and sign the digest.Finally,experiments showed good performance of robustness,tamper sensitivity and security,which meet the requirement of perceptual authentication.3.A robust and effective speech perceptual signature algorithm was proposed. Besides the performance of security and perception,the performance of efficiency was emphasized to meet with the real-time processing requirement.First,voiced frames were extracted to enhance the robustness,and reduce the computational complexity.Then,multi-band filtering and spectrum difference were used to extract the feature,and RM-EPN coding was used to compress the digest.Finally,MPKC was used to sign the result.This algorithm has rapid computing speed on the limited-resource platform.4.In order to improve the performance of MPKC,A series of optimizations were presented to deal with the main problems of MPKC realization.Matrix strategies for simplifying the operations on finite field L were investigated.A new calculation method of multiplication on field K based on bitslice and composite field was presented,A technique of sparse matrix compression was used to reduce the data storage space.Such methods were applied to SFLASH,and the results showed three times signing speed,three times verifying speed,and half of the storage cost comparing with the data of NESSIE.5.A speech perceptual authentication system was realized on FPGA platform with LEON3 processor and embedded Linux operating system.Experiments showed that,perceptual algorithms proposed in this paper could achieve the real-time authentication,and meet all the other requirements as well.