Autentikasi Biometrik Suara menggunakan ECAPA-TDNN untuk Peningkatan Keamanan Sistem Kendali Pagar Rumah

Abstract

Conventional home gate security systems relying on physical keys are vulnerable to loss and theft, whereas standard voice-controlled smart home systems focus solely on speech recognition without verifying the speaker's identity. This study aims to integrate voice biometric authentication into an IoT-based home gate control system to deliver dual-layer security. An experimental design utilizing a prototyping methodology was employed. The system implements the ECAPA-TDNN Deep Learning algorithm on a Raspberry Pi edge computing device to verify the homeowner's voice identity using Cosine Similarity, alongside an ESP32 microcontroller to drive a stepper motor via MQTT. Findings indicate high biometric accuracy with a False Acceptance Rate (FAR) of 3.75%, a False Rejection Rate (FRR) of 2.08%, and an Equal Error Rate (EER) of 2.92%, while all functional mechanical controls and live CCTV streaming executed reliably in real-time. Combining ECAPA-TDNN with IoT effectively establishes a secure, responsive gate control system independent of third-party cloud services. Future research is suggested to evaluate system performance in multi-user settings with diverse family voice profiles.