多特征和神经网络相结合的语音端点检测算法

摘要

为了提高语音端点检测正确率,提出一种基于多特征和神经网络相结合的语音端点检测算法.首先分别提取语音信号的短时能量特征、时域方差特征和频域方差特征,然后将这些特征量作为神经网络输入进行训练和建模,最后判断出该信号的类别.仿真实验表明,相对于单一特征语音端点检测算法,多特征融合和神经网络检测算法提高了语音端点检测正确率,具有更好的适应性和鲁棒性,对不同信噪比的信号都有较好的检测能力.%This paper presents a method for speech endpoint detection algorithm based on the combination of multiple features and neural network to improve the detection accuracy rate.Firstly,the features of short-time energy,time-domain variance and frequency-domain variance of speech signals are extracted respectively,and then these feature quantities are employed as the input of neural network for training and modelling,and finally the signal' s category are determined.Simulation experiments prove that compared with single feature speech endpoint detection algorithm,the proposed algorithm improves the detection accuracy rate,has better adaptability and robustness and has preferable detection ability on signals with different SNR.