摘要
使用模糊c均值聚类(FCM)结合贝叶斯准则(BIC)估计门限阈值时,FCM-BIC估计门限阈值算法(FBD)在低信噪比情况下鲁棒性较低,而且FCM对初始聚类中心敏感,容易陷入局部最优解。文章提出了一种基于遗传模拟退火(GASA)优化FCM-BIC算法的语音端点检测方法。该方法选用短时能量和谱熵作为门限参数,并融入了遗传模拟退火算法,将得到的聚类中心赋给FCM-BIC以确定信号特征的门限值,最后根据门限检测出语音端点。实验结果表明,选用谱熵作为门限参数在低信噪比噪声背景下鲁棒性更好;本文提出的方法的加权错误测度均小于传统双门限法(EZ)和FBD方法,在白噪声下算法改善效果更明显,在嘈杂噪声下端点检测效果最好。
When using fuzzy c-means clustering and Bayesian criterion to estimate the threshold threshold,FCM is sensitive to the initial clustering center and is easy to fall into the local optimal solution,and the FCM-BIC estimation threshold algorithm is low.The robustness is lower in the case of signal to noise ratio.This paper proposes a speech endpoint detection method based on genetic simulated annealing optimized FCM-BIC algorithm.The method uses short-time energy and spectral entropy as threshold parameters,and incorporates the genetic simulated annealing algorithm.The obtained clustering center is assigned to the FCM-BIC to determine the threshold value of the signal feature,and finally the speech endpoint is detected according to the threshold.The experimental results show that spectral entropy is more robust as a threshold parameter in the low SNR.The weighted error measure of the proposed method is smaller than the traditional double threshold method(EZ)and FBD method,and the algorithm is improved under white noise.The effect is more obvious,and the endpoint detection is best under noisy noise.
作者
汤琛
董胡
邹孝
马玉洁
钱盛友
TANG Chen;DONG Hu;ZOU Xiao;MA Yu-jie;QIAN Sheng-you(School of Physics and Electronics,Hunan Normal University,Changsha 410081,China)
出处
《电脑与信息技术》
2020年第3期8-12,共5页
Computer and Information Technology
基金
国家自然科学基金(项目编号:11774088)
湖南省自然科学基金(项目编号:2018JJ3557)
湖南省教育厅优秀青年项目(项目编号:17B025)。
关键词
端点检测
模糊C-均值聚类算法
贝叶斯信息准则
模拟退火算法
双门限法
endpoint detection
fuzzy C-means clustering algorithm
Bayesian information criterion
simulated annealing algorithm
double threshold method
