低信噪比环境下的语音端点检测算法研究

打开文本图片集

中图分类号：TP39；TN912.3 文献标识码：A 文章编号：2096-4706（2025）11-0005-04

‘，1，²(1.，浙江温州325003；2.温州技师学院，)

Research on Speech Endpoint Detection Algorithm in Low Signal-to-Noise Ratio Environment

FUHongzi',CHENXiduan1,ZHANGLin² (1.ZhejiangIndustryandTrade VocationalCollege,Wenzhou 325oo3,China; 2.Wenzhou TechnicianInstitute,Wenzhou 325ooo,China)

Abstract:To enhance the robustnessand accuracyof speech endpoint detection inlow Signal-to-Noise Ratio environment, a speech endpoint detectionalgorithmbasedonmult-windowspectralestimationand subbandquadratic spectral entropyis proposed.This algorithm integrates theadvantagesofsubbandquadratic spectralentropyand multi-windowspectral estimation. First,the quadraticpower spectrumof each frameofnoisyspeech signaliscomputed.Then,subbandanalysis is performed to derivethesubbandquadraticspectralentropy.Finalyontisbasisulti-windowspectralestimationofmulti-sine-windowis incorporated into thesubbandquadratic spectralentropytoformnew feature parameters fordetermining the startand end points ofspeechendpointdetection.ThroughMATLABsimulationverification,thisalgorithmhasbetternoiserobustessinlowSignalto-Noise Ratio environment compared to traditional speech endpoint detection algorithms.

Keywords:lowSignal-to-NoiseRatio;subbandquadraticspectralentropy;multi-windowspectral estimation;robustess

0 引言

语音端点检测是语音识别系统前端处理的关键环节，其检测结果的准确性直接影响语音识别系统的性能和效率。（剩余9093字）