A MATLAB-Based Time Delay Neural Network Approach to Speaker Recognition for Voice Biometric Attendance Systems: Model Development and GUI Design

Abstract

Traditional attendance systems often suffer from inefficiency, inaccuracy, and vulnerability to fraud. Utilization of voice biometrics can help to mitigate these downsides. This paper presents the development and design of a MATLAB-based voice biometric attendance system that is inspired by MathWorks' x-vector implementation. The system utilizes a Time Delay Neural Network (TDNN) for speaker recognition which has been trained on a 100 hour subset of the LibriSpeech dataset, with Mel-Frequency Cepstral Coefficients (MFCCs) being used as input features. A graphical user interface (GUI) has been developed within MATLAB App Designer which provides features such as user registration, login, voice enrollment, and attendance recording. The system also incorporates a local database for user enrollment data and employs text-dependent voice enrollment for enhanced user experience and accuracy. This work demonstrates the successful integration of the x-vector approach into a functional attendance system, highlighting its practicality and feasibility for real-world applications.