Design Of Access Control System Based On Speaker Recongnition

Speaker recognition technology is one of the computer security technologies developed in recent years. Speaker recognition, also named as voiceprint recognition, identifies different speakers through the personality traits, such as speaker's voice characteristics, speaking habits, etc. With the rapid development of society, it has been applied widely in many fields, like attendance system, security control and financial security. This paper studies on the text-independent speaker recognition algorithm and the access control system realized by the use of this algorithm on DSP. The speaker recognition algorithm in this paper makes the use of Mel-cepstrum and vector quantization technique.As for the selection of characteristic parameters, this paper uses the Mel-cepstrum, which takes full account of human ear's auditory property. As for the pattern matching techniques, this paper uses the vector quantization technique, a coding technique with high efficiency. When there is small amount of data used for training, the method based on vector quantization has big robustness than other methods. While, the method based on vector quantization is relatively simple and real-time. So far, speaker recognition technology based on vector quantization is still one of the most commonly used identification methods. This paper compares the speaker recognition methods in MATLAB environment between the one using dynamic time warping and the other one using vector quantization, and the result shows that for the text-independent speaker recognition, the vector quantization is better than dynamic time warping, so this paper uses the vector quantization as the pattern matching technique.As for the realization of access control system, the entire system consists of three parts, voice collection, speech signal storage and display of relevant information. The hardware is mainly based on the TI's digital signal processor TMS320VC5509, audio codec TLV320AIC23, off-chip SDRAM memory chip and LCD. The tasks based on the hardware include collecting voice signal through TLV320AIC23, transmitting the digital voice signal by McBSP to TMS320VC5509, accomplishing speech feature parameters extraction and identification algorithm on TMS320VC5509, displaying the final identification information through LCD, and sending corresponding control command.