基于RK3588的嵌入式声纹识别系统设计  

Design of Embedded Voiceprint Recognition System Based on RK3588

作　　者：陈德富[1] 周云龙 陈国军 宋京洋 林永权 CHEN Defu;ZHOU Yunlong;CHEN Guojun;SONG Jingyang;LIN Yongquan(Department of Information Engineering,Zhejiang University of Technology,Hangzhou,Zhejiang 310023,China;Zhejiang Wanxiang Precision Industry Co.,Ltd.,Hangzhou,Zhejiang 311215,China)

机构地区：[1]浙江工业大学信息工程学院,浙江杭州310023 [2]浙江万向精工有限公司,浙江杭州311215

出　　处：《自动化应用》2025年第6期18-21,共4页Automation Application

基　　金：杭州市重点科研计划项目(2024SZD1A19)。

摘　　要：为了推动声纹识别技术在实时性、准确性和便捷性方面的发展,设计了一种基于RK3588嵌入式开发板的声纹识别系统。该系统使用Pytorch深度学习框架在PC端搭建ResCNN网络进行模型的训练和推理测试,随后通过模型转换将训练好的模型部署至RK3588平台。界面设计采用了PyQT5,实现了直观的用户操作。实验结果表明,声纹比对推理的准确率达到96%,等错误率(EER)为3.30%,平均每组比对耗时约0.1 s。该系统满足实时性要求,适用于需要快速声纹验证的应用场景。In order to promote the development of voiceprint recognition technology in terms of real-time,accuracy and convenience,this paper designs a voiceprint recognition system based on the RK3588 embedded development board.The system uses the Pytorch deep learning framework to build a ResCNN network on the PC side for model training and reasoning testing,and then deploys the trained model to the RK3588 platform through model conversion.The interface design uses PyQT5 to achieve intuitive user operation.The experimental results show that the accuracy of voiceprint comparison reasoning reaches 96%,the Equal Error Rate(EER)is 3.30%,and the average comparison time per group is about 0.1 seconds.The system meets the real-time requirements and is suitable for application scenarios that require fast voiceprint verification.

关 键 词：声纹识别 RK3588 ResCNN 模型部署 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]