This demo showcases a voice-based multi-factor authentication system with two security layers: a Whisper-powered passphrase check using a randomly generated challenge phrase, and speaker identification using ECAPA-TDNN voice embeddings with an SVM classifier. Access is granted only if both checks pass. Even with the correct phrase, unregistered voices are rejected based on the classifier’s confidence score. The system is integrated with an ESP32-controlled smart door lock that unlocks after successful authentication and triggers alerts after repeated failed attempts.
Built with #Python, #FastAPI, #SpeechBrain, #Whisper, #scikit-learn, #Next.js, and #ESP32.
