Design And Implementation Of An Anti-recording Speaker Verification System

Password authentication is the most widely used authentication technology. It also has some defects because passwords are easy to be forgotten and stolen. Especially with the spurt of the Internet, users have to spend more time and energy on password memorization and management. Once one’s password is disclosed to others, he or she is more likely to suffer from greater loss. On the other hand, biometric authentication, using unique and stable biological characteristics for identification and needing no one to memorize complex passwords or take things with them, is safe and convenient.Speaker recognition, one of biometric techniques, is accurate, lowcost and non-contact. Nowadays, it has become hot in the field of personal identification. In many ways it has been applied, such as login system, telephone banking, judicial cases detecting, personalized search, etc. Yet, speaker recognition faces the problem that the audio may be replayed from a recorder in practical application. If the recognition system can not resist this kind of attacks, it will not satisfy security requirement. Considering this application scenario, this thesis designs and implements a speaker verification system with the ability to resist playback. The main work is as follows:1. Discuss theory and application of speaker recognition. Illustrate the meaning of the development of this anti-recording speaker verification system.2. Propose the scheme to build the system, that is using mandarin digit speech recognition(MDSR) to help build the system’s ability to resist the attack of playback.3. Design and establish speech corpus for the system.4. Design and implement the background system which is responsible for extracting feature, training and testing model, database operation, etc.5. Design and implement the client which interacts with the background system and provides a GUI for users to register and login.6. Evaluate the recognition performance of the system, including the efficiency and effect of speaker verification and MDSR. Demonstrate the system’s practical value through the assessment.