| In recent years,with the development of the embedded technology, speech recognition technology started from lab to market. Speech recognition technology is a kind of technology which make the machine convert the linguistic signals into the text or command through the process of recognition and understanding. It becomes a research focus because of its study difficulty and importance. Applying speech recognition technology in embedded systems makes embedded speech recognition technology become the new development direction in speech recognition field. And the application of embedded speech recognition technology has become a competitive high-tech emerging industry.In this paper,we studied the basic theory of speech recognition firstly,which includs the speech signal pretreatment, feature extraction and model matching,and the phonetic features parameters extraction method which based on MFCC+â–³MFCC is given. Based on the theory, we extracted 10 speech data which is from 0 to 9 of the specific person in laboratory environment. The experimental simulation of speech signal pretreatment method which include increase in advance ,endpoint detection and feature parameters extraction method which based on MFCC+â–³MFCC and voice on the simulation model matching algorithm was carried respectively.The effectiveness of the above theory was proved by the experimental result. Then,we applied the above research results to embedded speech recognition system. Based on the study of relevant theories of embedded speech recognition which include the system development environment,the choice of hardware processor and so on,we designed the hardware and software of embedded speech recognition system.The system includes hardware platform, Bootloader, Linux kernel and root filesystem. Hardware design part, we choose ARMS3C2440 processor as core development board, then designed the hardware modules including system control , system audio input/output , system storage and communication interface,and build the system hardware platform, through the serial port UDA1341TS pronunciation chip to realize the software of speech signal A/D sampling. The selection of ARMS3C2440 greatly improved the system speed and performance. Software design part, the system software development process and implementation scheme is introduced. The phonetic features parameters extraction method which based on MFCC+â–³MFCC and DTW algorithm functional modules are debugged and implementated in Linux development environment.Finally,through the ARMS3C2440 collect speech data and the specific experimental results verified the performance in specific person isolated words of speech recognition system which based on MFCC+â–³MFCC phonetic features parameters extraction method and the DTW template matching algorithm, system test results to achieve the desired goals. |
