An Overview of Advancements in Lie Detection Technology in Speech

Introduction

Lie detection is a problem associated with psychological calculation (Hartwig & Bond, 2011; Levitan et al., 2015; Bond & DePaulo, 2006; Elad & Ben Shakhar, 2006), which aims to judge the psychological state of an individual by analyzing the characteristic information of their lying speech signals. The psychological state of lying is complex and can be influenced by the interaction between emotion, cognition, and willpower (Ekman et al., 1991; Chen et al., 2005; Zhao et al., 2004; Nitsch & Popp, 2014). Therefore, the influence of various factors must be comprehensively considered in the research process. The technology of lie detection employs physiological indicators and signs, such as the physiological parameters of the brain image, electroencephalogram (EEG), electromyography (EMG), heart rate, and voice. Considering this, the psychological states of lying can be measured, explained, and quantitatively evaluated (Liu & Li, 2013; James et al., 2009). In 1991, with the support of the Ministry of Public Security, the Chinese Academy of Sciences independently designed and developed China’s first lie detector called PG-1. This polygraph uses voice, skin electricity, and breathing parameters as test parameters. In 2006, Professor Yang developed the event-related potential (ERP) lie detector. This detector was the first lie detector with independent intellectual property rights in China. This successfully filled a gap in the field of lie detection research in China (Yang & Liu, 2016). Currently, the lie detection method of using a polygraph is common worldwide. In this method, the testers are asked to answer pre-set questions related or unrelated to the case, and there are psychological and physical changes when the tester lies (Meng & Zeng, 2000; DePaulo et al., 2003). A recent study found that brain waves can change and that the liar cannot be controlled when a person lies. This is the brain ERP that records changes in brain waves every millisecond. Therefore, ERP, which directly records changes in EEG, has become the third generation of lie detection technology. Functional magnetic resonance imaging (fMRI) detects lies by observing changes in oxygenated hemoglobin levels in brain regions. The amount of oxygenated hemoglobin in the brain region was detected by increasing the level of activation in the brain region.

Artificial intelligence is well developed, and lie detection technology in speech has gained attention, making it a prevalent topic in the field of speech research (Li et al., 2017; Gou & Wang, 2012; Chao et al., 2020), such as emotion recognition (Perez-Gasper et al., 2016; Nitsch & Popp, 2014; Wang et al., 2010; Takehara et al., 2016; Sanchez-Valdes & Trivino, 2015), psychological recognition (Qi, 2013; Gamer et al., 2006), and human–computer interaction (Das et al., 2015; Han et al., 2016). Through the normal measurement of speech features, the psychological state of the testers can be objectively and effectively analyzed. In lie detection research, several studies have combined psychological computing and achieved positive progress. Existing methods of speech lie detection include physiological information, natural language processing, and acoustic feature parameter detection (Alonso et al., 2015; Hamilton, 2005; Hu et al., 2012; Zhang et al., 2002). These are explained as follows (Pan, 2016).