Speech Deception Detection Algorithm Based on SVM and Acoustic Features
Huawei Tao, Peizhi Lei, Mengzhe Wang, Jie Wang, Hongliang Fu · 2019
In order to make full use of the advantages of speech-based deception detection, such as high concealment, low cost and easy operation, a speech-based deception detection algorithm based on Support Vector Machine (SVM) and acoustic features (AF-SVM) is proposed. Firstly, the deception corpus containing 388 speech data is constructed. Then, the zero-crossing rate, pitch rate, short-term energy and, Mel frequency cepstral coefficients of speech and other features are extracted and normalized. Finally, SVM classifier is used to train and classify the acquired feature data. The experimental results on corpus show that the recognition accuracy can reach more than 80%, and the proposed algorithm can effectively detect deception.