Research Of Text Dependent Speaker Recogniton On Embedded System

Speaker recognition is the computer task of validating a user's claimed identity using characteristics extract from their voice. Along with the development of research in speaker recognition theory, application of speaker recognition has been developed and used in many fields including telecommunication, security banking etc. Implementation of the algorithm on embedded systems is the key step of transforming speaker recognition theory to application. A system of text-depended speaker recognition is implemented on TDS320DM642 EVM board in this paper.The state of speaker recognition application is discussed fist, then key algorithms of speaker recognition such as feature extraction, modeling, and decision method is introduced. After comparison of calculation intense, and recognize result of different method, a system suitable for embedded is implemented through MATLAB, and the viability of the system is verified.A whole solution of embedded speaker recognition system is proposed after simulation and comparison of algorithms. Most work is on porting the system to DSP based board from PC, programming on DSP/BIOS which is a real time operation system, implementing fixed point speaker recognition.Finally, optimizing of the system is implemented. To meet the requirement of real time embedded system, optimizing programming based on specific hard ware is implemented. Several time consuming algorithms such a feature extraction, modeling, is optimized. At last, experiment based on real time data is proceeded, the result shows that requirements of real time is meted, and the recognize rate of speaker is 90%.