Identifying Duress Through Voice During Speaker Authentication

Biometric authentication improved accessibility to sensitive information as it provided a faster and more effortless approach to accessing sensitive data. The accuracy of many biometric authentication approaches was enhanced to the extent that it has become implemented as primary authentication. Regrettably, a user can still be authenticated under duress. This research suggests an approach for detecting duress when a person authenticates themselves using their voice. To detect possible duress within the RAVDESS dataset, several combinations of feature extraction techniques (MFCC, Chroma, Mel, Contrast, and Tonnetz) were employed in conjunction with the Random Forest classifier. As a result, a modified confusion matrix was used to more accurately measure performance and provide context for the results of emotion detection.