Design and implementation of a high-performance client/server voiceprint recognition system

• Published in: 2012 IEEE International Conference on Information and Automation pp. 704 - 708
• Main Authors: Gao Guanyu, Kang Kai, Guan Sheng Xiao, Gao Guanhua
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01-06-2012
• Subjects: client/server; endpoint detection; Feature extraction; Instruction sets; multi-thread; parallel algorithm; Parallel algorithms; Servers; Spectrogram; Speech recognition; Vectors; voiceprint recognition
• ISBN: 9781467322386; 1467322385
• DOI: 10.1109/ICInfA.2012.6246767

This paper designs and implements an authentication system of client/server architecture based on voiceprint recognition. The client uses MFCC method to extract feature vectors from the speaker's voice and the server uses VQ method for recognition. We propose a new method of endpoint detection named short-term variance which can effectively pick up the voice signal from the original signal in actual environment. With the improved endpoint detection algorithm, the system can effectively resist the noise and achieve a higher recognition rate. In order to boost the server's performance, we implemented a new parallel algorithm for VQ codeword search on the SMP system. By using this method, the server improved the processor load rate and the speed of operation, as well as reduced the system response time. In the experiment, we evaluated the system recognition accuracy and efficiency of the VQ parallel algorithms.