Design And Implementation Of Speech Recognition System Based On FPGA

Voice interaction has become a hot research field in recent years.Smart phones,car electronics and other devices with voice interaction has become a trend.Voice interaction simplifies the way of human-computer interaction,and brings a good user experience.In the research of speech recognition,the speech recognition algorithms have achieved outstanding results,but the research progress in hardware platform is relatively slow.The existing speech recognition platform has many disadvantages,such as complex structure,high power consumption and high cost.With the development of microelectronic technology,high performance and low power FPGA(Field Programmable Gate Array,field programmable gate array)constantly,FPGA a new generation of integrated central processor and a digital processor core for the realization of the system on chip(SoC,System on Chip)provides a powerful support.Therefore,this topic from the SoC architecture design,hardware and other aspects of the algorithm to provide a solution for the voice recognition system products.In this paper,a speech recognition algorithm based on Hidden Markov model and mixed Gauss model is studied.First,this paper analyzes the performance of speech recognition system,based on Xilinx's Kintex-7 FPGA as the main control chip and the embedded AndesCore N10 as CPU SoC architecture design,and designed the FPGA development board.Secondly,in logic design,focus on solving the floating-point SoC hardware architecture design,algorithm of fixed point optimization and high data throughput of system memory cache design bottleneck,sub module is the hardware design and system validation and evaluation scheme etc..Using Verilog hardware description language to complete the behavioral description of speech recognition system,to achieve a set from the voice signal acquisition,data caching,data model recognition processing,the results show that the complete demo system on FPGA development board.Finally,using Verdi simulation platform and online logic analyzer(ChipScope),the system is simulated and tested.The results of FPGA data test show that the timing of each module of the speech recognition system is correct,the speech recognition rate is 95%,and the speech recognition time is about 10 ms.On the basis of the FPGA prototype verification,it lays the foundation for the next step of the speech recognition algorithm.