面向语音谎言检测的智能机器学习特征融合算法研究
投稿时间:2022-06-20  修订日期:2022-09-06  点此下载全文
引用本文:
摘要点击次数: 18
全文下载次数: 0
作者单位邮编
林丽 陕西警官职业学院 710021
基金项目:陕西省教育厅专项科研计划项目:深层语音情感分析系统在侦查讯问中的应用 项目号20JK0065
中文摘要:摘要:针对目前语音谎言检测识别效果低、特征提取不充分等问题,提出了一种基于注意力机制的欺骗语音识别网络。首先,将双向长短时记忆与帧级声学特征相结合,其中帧级声学特征的维数随语音长度的变化而变化,从而有效提取声学特征。其次,采用基于间注意增强的卷积双向长短时记忆模型作为分类算法,使分类器能够从输入中学习与任务相关的深层信息,提高识别性能。最后,采用跳跃连接机制将时间注意增强卷积双向长短时记忆模型的底层输出直接连接到全连接层,从而充分利用了学习到的特征,避免了消失梯度问题。实验阶段,与LSTM以及其他基准模型进行对比,所提模型性能最优。仿真结果进一步验证了所提模型对语音谎言检测领域发展及提升识别率提供了一定借鉴作用。
中文关键词:关键字:谎言侦查  声学特征  特征提取  注意力机制  长短时记忆网络
 
Research on intelligent machine learning feature fusion algorithm for speech lie detection
Abstract:Abstract: A deceptive speech recognition network based on attention mechanism is proposed to solve the problems of low performance of speech lie detection and recognition and insufficient feature extraction. Firstly, the bidirectional long-term and short-term memory is combined with frame level acoustic features, in which the dimension of frame level acoustic features changes with the change of speech length, so as to effectively extract acoustic features. Secondly, the convolution bidirectional long-term and short-term memory model based on inter attention enhancement is used as the classification algorithm, so that the classifier can learn the deep information related to the task from the input, and improve the recognition performance. Finally, the jump connection mechanism is used to connect the bottom output of the time attention enhanced convolution bidirectional long and short-term memory model directly to the full connection layer, so as to make full use of the learned features and avoid the vanishing gradient problem. In the experimental stage, compared with LSTM and other benchmark models, the performance of the proposed model is the best. The simulation results further verify that the proposed model can provide a reference for the development of speech lie detection and improve the recognition rate.
keywords:Keywords: Lie detection  Acoustic characteristics  Feature extraction  Attention mechanism  Long and short term memory network
查看全文   查看/发表评论   下载pdf阅读器