A Lie Detection Method Based on Deep Recursive Conditionally Restricted Boltzmann Machine

Abstract

The invention discloses a lie detection method based on a depth recursion-type condition-restricted Boltzmann machine. Firstly, in a continuous voice paragraph, by using the fact that the condition-restricted Boltzmann machine has good modeling features and a simple reasoning process on a time series, a training sample is modelled to obtain high-order statistic information about whether or not a speaker lies; then, supervised parameter training is conducted on a recurrent neural network by using the high-order statistic information and the label of the training sample. After initialization parameters of the condition-restricted Boltzmann machine and the recurrent neural network are obtained, the two basic network units are constructed from bottom to top; on a validation data set, the parameters of the recurrent neural network are finely adjusted based on least square regression; by using the constructed networks, voice signal features of the speaker are tested. According to the method,the result of lie detection can be obtained automatically, and the method has a relatively-high recognition rate, low requirements on the professional knowledge and skills of evaluators and high testefficiency.

Description

technical field [0001] The invention relates to a speech polygraph technology, in particular to a method for polygraph detection by utilizing the speech information of the speaker's context. Background technique [0002] The basic principle of "polygraph detection" is that the psychological changes of people when lying will inevitably cause changes in some physiological parameters (such as skin electricity, heartbeat, blood pressure, breathing brain waves, voice), usually it is only restricted by autonomic nerves and difficult to be affected by brain consciousness. control. Therefore, the traditional polygraph technology is a combination of psychology and physiology and other disciplines, and detects the intention and state of the individual's inner concealment through the electrophysiological parameter testing system. At present, a large amount of psychological work is based on facial expressions, physiological activities and gestures as test clues for lies. There are thr...

AI Technical Summary

Problems solved by technology

[0005] Purpose of the invention: In order to address the deficiencies of the prior art and solve the problems of excessive cost and deviation in the use of empirical information for lie detection in the prior art, the present invention proposes a method based on a deep recursive conditionally restricted Boltzmann machine. polygraph method

Images