Research area
Face and voice authentication, automated attendance and exam invigilation.
Identity checks are needed everywhere from building access to examination halls, yet passwords are easily shared and single biometrics can be spoofed. We look for systems that are secure enough to trust and light enough to run on inexpensive hardware.
Our systems combine face recognition and voice verification, and use lightweight networks for edge deployment. In the two-step authentication design, a VGG16 face classifier identifies the user and a ResCNN speaker encoder then verifies the voice; both steps must pass. The invigilation system pairs Swin-Transformer semantic segmentation with ShuffleNet V2 face identification to detect cheating behavior.
Representative projects, all on GitHub, are Two-Step Multi-Biometric Authentication (ICETA 2024), the Face Recognition Attendance Management System and Invigilation Using Lightweight Semantic Analysis.
A face match picks the user's enrolled voice reference, and speaker verification must then pass as well; either failed check denies access. In the code release, face identification reaches about 95% accuracy on five enrolled students and speaker verification about 99% on LibriSpeech, each tested on its own. Figure from the code release.
View on GitHub
Three models look at each camera frame: who the student is, whether a phone or earphones are in view, and where each body is. Four checks then decide whether to warn the invigilator. The models are small enough for edge devices. Redrawn from the flow diagram in the repository.
View on GitHubLets you in only when both your face and your voice match, with an ordinary webcam and microphone.
A two-step authentication design in which a VGG16 face classifier identifies an enrolled user and a ResCNN speaker encoder then verifies their voice, using only an ordinary webcam and microphone; published at ICETA 2024.
People Kuan-Wei Chen, Ting-Yi Lin
A camera helps proctor exams: it recognizes each examinee's face and understands the scene, on hardware light enough for the classroom.
Vision-based exam invigilation that combines Swin-T semantic segmentation (92.74% accuracy) with ShuffleNet V2 face identification (98.58%), fused at decision level for edge deployment.
People Ting-Yi Lin
Takes the roll by recognizing faces through a webcam.
A Tkinter and OpenCV desktop app that captures face images, trains a recognizer, and logs attendance in real time to CSV files or a MySQL database, with manual entry as a fallback. Supported by TEEP.
2025 IEEE International Conference on Consumer Electronics - Taiwan (ICCE-Taiwan), pp. 349–350
Compares the speech of Taiwanese Mandarin–English bilinguals with Indian English speakers using MFCCs, pitch contours, spectrograms and wavelet analysis: Taiwanese speech shows more pitch variation and tonal shifts, Indian speech a steadier rhythm. The findings guide the choice of speech-recognition and speaker-verification models for diverse speakers.
Two-step Authentication: Multi-biometric System Using Voice and Facial Recognition
IET International Conference on Engineering Technologies and Applications (ICETA 2024), IET Conference Proceedings, vol. 2024, no. 22, pp. 11–12
Combines face and voice recognition into a low-cost authentication system for a small group that needs only a camera and a microphone: a face model identifies the user, then a speaker model verifies the voice. Code and models are public.
Interested? Visit the lab or write to us.