Research And Implementation Of Voice Authentication Based On Smart Phone

In recent years, with the rapid advances in smart phone technology, gave birth to the vigorous development of mobile commerce. At the same time, security and authentication issues of mobile commerce based on smartphone platform are also prominent, which has become the key to the further development of mobile commerce. At present, preset symbols keys are mostly used to authenticate, but the security of authentication by this method is mainly relied on the complexity of the key symbols. However the complexity of the key and ease of use are irreconcilable contradictions. On the other hand, with the further development of biotechnology, a variety of biometric authentication technologies have been more and more widely used. As biometrics in one, speaker recognition technology is easy to apply to the existing smart phones, and there are some advantages compared with other cryptographic techniques. Therefore, we consider the speaker recognition technology to Android system security certification application and the purpose of the application is used as a low-intensity password system, or as one of factors in a multi-factor authentication.This application system uses a B/S architecture. The client is based on the Android system and acts as the speaker’s voice acquisition, account management, speaker feature computing functions. The main function of the server is to complete the training and recognition main speaker model. Moreover, the entire application system has been optimized to local conditions based on the features of the Android platform, specifically as follows:1) Proposed the efficient parallel algorithms of speaker characteristic parameters MFCCBecause speaker feature (MFCC) extraction are mutually independent and most smart phones are multi-core, so this thesis presents an efficient parallel algorithm of MFCC. When the system is up and confirms that the current device hardware environment, the system will obtain the results of one or multi-core, and then generate one or more MFCC parameter extraction service threads for parallel operation, which speed up the feature extraction process and improve the experience level of the users.2) Optimization method of GMM modelIn this thesis, as a strong scalable clustering algorithm, K-Means is used in this system based on the performance limitations of the phone and specific applications for real-time requirements and some fast algorithms are presented for GMM model optimization. Meanwhile, in order to improve the accuracy of speaker identification, the system is designed to run on a one hundred prefabricated random imposters which is used to determine the threshold for speaker identification.3) System design and implementationAndroid system software architecture is studied in this thesis, and the overall architecture of the speaker authentication system is presented based on the Android environment. Then the overall system design, module design, database design and interface design are completed. Ultimately the system was implemented on the Acer Liquid E600 experiment platform and basically achieved the goal of the system design.