基于FPGA的声纹识别系统设计  

Speaker Recognition System Based on FPGA

作　　者：贾子龙 潘士杰 郭子昊 唐进[1] 姚燕[1] JIA Zilong;PAN Shijie;GUO Zihao;TANG Jin;YAO Yan(School of Automation,Beijing University of Posts and Telecommunications,Beijing 100876,China)

机构地区：[1]北京邮电大学自动化学院,北京100876

出　　处：《电子器件》2022年第4期997-1003,共7页Chinese Journal of Electron Devices

摘　　要：随着生物识别技术在各行各业之中开始普遍使用,作为人类最高频的交互方式,声纹识别成为生物识别技术中一种不可替代的解决方案。本文设计了一种基于现场可编程门阵列(Field Programmable Gate Array,FPGA)的声纹识别系统。该系统基于MFCC提取的声音特征通过卷积神经网络模型进行声纹识别,并结合IP核对卷积运算进行加速,试验测试表明,该系统可以充分发挥FPGA的高密度、高效率优势,提高CNN的运行效率、优化其前馈网络结构,从而实现更快速、更精准的声纹识别。As the most common way of human-computer interaction,speaker recognition has become an irreplaceable solution in biometric technology.A speaker recognition system is designed based on field programmable gate array(FPGA).The system aims to achieve speaker recognition effect by putting the MFCC features into a convolutional neural network(CNN)model,and tries to accelerate the convolution related calculation with the help of self-designed IP cores.The results show that the system can give full play to the advantages of high density and efficiency of FPGA,improve the efficiency of CNN,and optimize the structure of the feedforward network to achieve a faster and more accurate speaker recognition.

关 键 词：声纹识别 MFCC 卷积神经网络 FPGA硬件加速 

分 类 号：TP331.2[自动化与计算机技术—计算机系统结构]