Abstract: In order to improve detection rate of the speech endpoint, this paper proposes a speech endpoint detection method based on BP neural network optimized by improved momentum particle swarm optimization algorithm. The features of speech signals are extracted by wavelet analysis, then the features are input to BP neural network to build the speech endpoints detection model in which the BP neural network’s parameters are optimized by particle swarm optimization algorithm, the simulation experiments are carried out on Matlab environments. The experimental results show that the proposed method improves the detection rate, and reduces the false detection rate and false negative rate effectively, WA-IMPSO-BP is a high detection rate and strong resistant noise performance speech detection algorithm.

Key words: wavelet analysis, neural network, speech endpoints, particle swarm optimization algorithm, feature selection

摘要： 为了提高语音端点检测率，提出一种改进动量粒子群优化神经网络的语音端点检测算法（WA-IMPSO-BP）。利用小波分析提取语音信号的特征量，将特征向量作为BP神经网络输入进行学习，并采用粒子群算法优化BP神经网络参数，建立语音端检测模型，在Matlab环境下进行仿真实验。仿真结果表明，WA-IMPSO-BP提高了语音端点检测率，有效降低了虚检率和漏检率，表示WA-IMPSO-BP是一种检测率高，抗噪性能强的语音检测算法。

关键词: 小波分析, 神经网络, 语音端点, 粒子群优化算法, 特征选择