Design And Implementation Of Isolated Word Speech Recognition System Based On FPGA

In recent years, the study of speech recognition technology concentrated mainly in the direction of algorithm improvement and software implementation, and has made a wealth of research results. However, the progress was very slow in the hardware implementation. The existing products of speech recognition are difficult to meet the requirements of market for convenient, energy saving, real-time and low cost. With the rapid development of microelectronics technology and theoretical in-depth study, the speech recognition system will gradually become more practical, smaller, which makes speech recognition system based on hardware to become a hot research field of speech processing.In the first part of this paper, it used MATLAB software to simulate each part of the algorithm in the isolated word speech recognition system. Based on that, this paper applied Xilinx’s Viterx II Pro development board as a platform to research. Then, it used ISE10.1that was an integrated development environment to achieve the isolated word speech recognition system which based on FPGA. It also used the online logic ChipScope Analyzer to verify the hardware capabilities. In this system, it flexibility used a variety of skills that based on the design of FPGA hardware and take advantage of the high-speed parallel processing superiority that the FPGA had. These made the capability of digital signal processing to have a great improvement on the isolated word speech recognition system.The research of this paper contains the following aspects:(1) The principle of speech recognition system is studied, the soft flow diagram is given, and hardware frame of isolated word speech recognition system is designed. Meanwhile, based on the analysis of resource in Viterx II Pro chip, this paper determined the overall design of isolated word speech recognition system which based on FPGA.(2) The speech recognition system principle is analyzed and studied, such as speech signal pretreatment, endpoint detection, feature extraction and pattern matching module, then did functional simulation for each module based on MATLAB. Meanwhile, in this paper, the coefficients of MEL filter bank, the best codebook of VQ and the template parameters of HMM are extracted.(3) In this article, the isolated word speech recognition system uses a modular design approach that uses hardware description language (Verilog HDL) to complete the circuit design of the isolated word speech recognition system, such as speech endpoint detection algorithm based on energy, feature extraction based on MFCC, vector quantization algorithm and pattern matching algorithm based on HMM. Meanwhile, each module has been achieved on the hardware. And then test and verified each module via ChipScope Analyze. At last, this paper combined all modules into isolated word speech recognition system.(4) In the last part of this paper, it has done the overall system test and analyzed the performance and real-time of isolated word speech recognition system. According to analyzed this system we can know that it has a good stability and real-time to deal with the target.