无声语音识别中的端点检测算法  

Endpoint Detection Algorithm in Silent Speech Recognition

作　　者：彭佩瑶 杜月山 韦峻峰 PENG Peiyao;DU Yueshan;WEI Junfeng(School of Electronic Information,GUANGXI MINZU UNIVERSITY,Nanning 530006,China)

机构地区：[1]广西民族大学电子信息学院,广西南宁530006

出　　处：《电声技术》2022年第11期139-141,156,共4页Audio Engineering

摘　　要：无声语音识别常以表面肌电信号作为研究对象。表面肌电信号的端点检测是影响识别结果的一个重要因素。表面肌电信号与语音信号有类似之处。借助语音端点检测的方法对表面肌电信号进行分割是一种可行的思路。基于此,采用子带谱熵和梅尔倒谱距离作为信号端点检测的判决依据,通过粒子群算法优化支持向量机分类器给出端点检测结果。结果表明,在不同信噪比条件下,该算法有最高的检出率与最低的错误率。对于基于K最近邻(K-Nearest Neighbor,KNN)的无声语音识别任务,识别率达95.3%。Surface electromyographic signal is often used as the research object for silent speech recognition.The detection of surface electromyographic signal endpoint is an important factor affecting the recognition results.Surface electromyographic signal is similar to speech signal.It is a feasible idea to segment surface electromyographic signal by speech endpoint detection.In this paper,subband spectrum entropy and Mel cepstrum distance are used as the decision basis for signal endpoint detection,and support vector machine classifier is optimized by particle swarm optimization algorithm to give the endpoint detection results.The results show that the algorithm has the highest detection rate and the lowest error rate under different signal noise ratio conditions.For the silent speech recognition task based on K-Nearest Neighbor(KNN),the recognition rate is 95.3%.

关 键 词：无声语音识别 表面肌电信号 子带谱熵 梅尔倒谱距离 

分 类 号：TN911.7[电子电信—通信与信息系统]