智能家居声控信号声纹识别算法  

Voiceprint recognition algorithm of smart home voice control signal

作　　者：朱敏[1] ZHU Min(School of Electrical and Electronic Engineering,Anhui Sanlian University,Hefei 230601,China)

机构地区：[1]安徽三联学院电子电气工程学院,安徽合肥230601

出　　处：《高师理科学刊》2023年第5期29-33,共5页Journal of Science of Teachers＇College and University

基　　金：安徽省自然科学重点研究项目(KJ2021A1190)——基于声纹识别的智能家居声控信号语音识别算法的研究。

摘　　要：为了解决智能家居语音环境中的安全性问题,提出了一种基于隐马尔可夫模型-通用背景模型(HMM-UBM)的声纹识别算法,实现说话人的准确有效识别.对智能家居语音信号进行预处理,采用基于能量和过零率的双门限比较法进行端点检测,通过MFCC特征提取,进行说话人的HMM-UBM模型搭建与训练,然后进行测试.该方法解决了在声纹训练过程中数据量不足的问题.通过Matlab2016a进行算法的仿真,将测试结果与传统高斯混合模型-通用背景模型(GMM-UBM)测试结果对比分析.结果表明,本系统识别正确率高于传统模型,实现了智能家居语音识别系统的安全可靠应用.Considering the security of speech environment in smart home,a voiceprint recognition algorithm based on hidden Markov model-universal background model(HMM-UBM)was proposed to achieve accurate speaker recognition.The speech signal of smart home is preprocessed,and the endpoint detection is carried out by the double-threshold comparison method based on energy and zero crossing rate.The HMM-UBM model of speaker is built and trained by MFCC feature extraction,and then the test is carried out.This method solves the problem of insufficient data volume during voiceprint training.The algorithm is simulated through Matlab2016a,the test results were compared and analyzed with the traditional Gaussian mixture model-universal background model(GMM-UBM),the results show that this system has a higher recognition accuracy than traditional models,achieving safe and reliable application of smart home speech recognition system.

关 键 词：声纹识别 智能家居 隐马尔可夫模型 通用背景模型 MFCC特征提取 

分 类 号：TP29[自动化与计算机技术—检测技术与自动化装置]