Deep Neural Network Architectures for Speech Deception Detection: A Brief Survey

The task of detecting deception has a long history since using the polygraph. In contrast, spot deception in conversational speech has been proved to be a current complex challenge. The use of this technology can be applied in many fields such as security, cybersecurity, human resources, psychology, media, and also for suspect interrogation. Due to the difficulty of detecting lies through speech, many approaches are applying deep learning combining audio of speech and textual characteristics from audio transcription. Many techniques have been developed to spot deceit through speech, and the purpose of this paper is to discuss in more detail these approaches. We discuss deep learning-based techniques and also other aspects such as available datasets and metrics. Finally, we argue about the limitations and examine promising and future works.