Researches And Implementation On Speaker Recognition Algorithms And Systems

Speaker recognition is the technique of automatically recognizing who is speaking by using speaker's specific information included in speech signal. With the development of communication and information technology, it is getting more and more attention for its bright future. At present, most speaker recognition systems are working on the computer. After primarily focusing on technique of feature extraction and recognition, this paper realizes a set of practical speaker verification system on SPCE061A, which can be applied to many fields of voice control.The paper's main works are as follow:(1) Several methods of getting pitch period are researched in details. According to the real-time demand for the systems, the methods for pitch searching are simplified as follow: after searching the peak value point every third point within pitch range, search the peak value point every point near the current peak value point.(2) Two kinds of representative features, linear prediction cepstrum coefficient (LPCC) and mel-frequency cepstrum coefficient (MFCC), are Analyzed and researched. The performances of LPCC and MFCC are compared respectively on SPCE061A or computer platform. Experiments indicate that the systems can gain better performances adopting LPCC than adopting MFCC in SPCE061A platform.(3) Dynamic Time Warping (DTW) algorithm is researched in details. Because of limited memory resource in SPCE061A, the whole section's DTW algorithm is simplified to segment DTW algorithm. The segment DTW algorithm reduces the memory spending for the matching paths. In addition, early termination technique is adopted, thus the calculating amounts are greatly reduced.(4) Because the speaker's status is changing slowly all the time, a new online pattern updating technique is presented. After optimizing the mode of'verifying after training'to the mode of'training while verifying after training first', the speaker verification systems have stronger robustness.(5) Programming with assembler language, a set of practical speaker verification system on SPCE061A is realized.The noise robustness is one of the key problems for the practicability of speaker recognition systems. How to keep performance of the systems not dropping in a situation that the noise environment changes, further research and practice is necessary. With the development of relevant subjects, some more practical, more high-powered speaker recognition systems will appear and apply to people's actual life extensively.