Personal Threshold in a Small Scale Text-Dependent Speaker Recognition

In this research, we implement a biometric authentication system on the Android platform, which is based on a text-dependent speaker recognition. The application makes use of the Modular Audio Recognition Framework (MARF), from which some of the algorithms are adapted in the pre-processing and feature extraction. In addition, we employ the Dynamic Time Warping (DTW) algorithm in the training and comparison processes to create reliable feature references and generate countable distance information. In particular, we introduce the personal thresholds to further optimize the system performance for each individual user. Our experimental results confirm that one can significantly improve the system performance in terms of the False Acceptance Rate (FAR) and False Rejection Rate (FRR), through using a collective training procedure and the personal thresholds.