Distinguishing deception from non-deception in Chinese speech

Deception detection is becoming indispensable to a growing number of applications in law enforcement and other government agencies. Recently, many researchers from both speech signal area and machine learning area have already shown that automatically deception detection from speech is promising. While there are a large amount of research works on English deception detection, few efforts have been put on Chinese which is quite different due to the culture divergence. In order to show the full potential of automatically deception detection, in this paper, we first construct the deceptive and non-deceptive Chinese speech corpus which has not been published so far. And then we propose a novel machine learning-based approach to detect deception in the same gender. Several popular machine learning algorithms are applied. Moreover, a transfer learning-based algorithm is applied for cross-gender deception detection. Experimental results show that our approach performs well on real-world corpus. In intra-gender deception detection, our approach can achieve roughly the same accuracy as the traditional method on English corpus. This means our corpus is reasonable and can be used for deception detection research. In cross-gender deception detection, our approach also outperforms the baseline methods.