|Title:||A Text Dependent Speaker Recognition Using Vector Quantization|
|Keywords:||Speaker Recognition, Vector Quantization, Acoustic Vector, Codebook, LBG Algorithm|
Nowadays, security concern is an important issue in keeping own identity secret to others, allowing only specified person to access, storing user database and so on. Speech based identification is used to verify the person’s identity for controlling access to services such as database access, banking by telephone, voice dialing, telephone shopping, information services, voice mail, security control for confidential information areas, and remote access to computers. Speech is used for identification of a human as the characteristics of vocal cord are different in each individual. On the basis of discriminatory information in speech waves, a specific speaker can be identified. In this paper, a microcomputer based speech recognition system has been designed using Vector Quantization (VQ) as a basis for identification. All speakers were modeled by a codebook of 32 vectors using LBG (Linde, Buzo and Gray) splitting algorithm. The speakers were prompted to say their nick name, different names for different speaker that is why it is called text dependent speaker recognition. Recognition decision is taken on the basis of the lower VQ distortion value with database speech and new samples of register user. If register user give the voice then the system measures VQ distortion values and based on lower VQ distortion value identification is done. Unknown speaker is discarded, based on comparison with threshold of each database speaker. We able to get 91.67% recognition rate with l2 database speaker. These recognition rate decreases as the numbers of speaker increase as decisions are made on comparisons of various VQ distortion data- causes’ greater chance to make mistake.