SOM-LSTM递归神经网络语音端点检测系统  

SOM-LSTM Recursive Neural Network Speech Recognition System

作　　者：唐铠 陆鹏 Tang Kai;Lu Peng(College of Computer and Information Technology, China Three Gorges University Yichang 443002 China;School of Mechanical Engineering, Yancheng Institute of Technology, Yancheng 224000, China)

机构地区：[1]三峡大学计算机与信息学院,湖北宜昌443002 [2]盐城工学院机械优集学院,江苏盐城224000

出　　处：《信息通信》2019年第5期50-53,共4页Information & Communications

摘　　要：语音端点检测是语音信号处理的一个重要环节,传统的语音端点检测算法往往是基于短时能量以及过零率等实现,在低信噪比的环境下,检测算法的准确度较低。因此,提出了一种基于自组织映射(SOM)神经网络和长短时记忆(LSTM)递归神经网络相结合的端点检测算法。该算法通过检测语音信号在每个时间节点上的特征属性利用SOM神经网络进行聚类,并根据每个时间节点的语音状态对聚类结果进行调整,构造能够判别语音状态和噪声状态的样本作为LSTM递归神经网络的输入,利用LSTM递归神经网络实现端点检测的目的。Speech endpoint detection is an important part of speech signal processing. The traditional speech endpoint detection algorithm is based on short-term energy and zero-crossing rate. In the low SNR environment, the accuracy of the detection algorithm is low. Therefore, an endpoint detection algorithm based on self-organizing map (SOM) neural network and long-term short-term memory (LSTM) recurrent neural network is proposed. The algorithm uses the SOM neural network to detect the feature attributes of the speech signal at each time node, and adjusts the clustering result according to the speech state of each time node to construct a sample that can discriminate the speech state and the noise state. Input to the LSTM recurrent neural network.

关 键 词：语音端点检测 长短时记忆递归神经网络 自组织映射神经网络 语音状态堆叠 无监督学习 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]