Voice In Ear: Spoofing-Resistant and Passphrase-Independent Body Sound Authentication
summary
System authenticates users by measuring speaking-induced body sound in the ear canal and exploiting left/right asymmetry, claiming passphrase-independence and spoof resistance. It maps air-to-in-ear signals with cycleGAN, estimates ear transfer functions for asymmetry, and classifies with a GMM-UBM i-vector pipeline plus an ensemble tree fused by soft voting. Prototype evaluation on 23 people achieves 3.64% EER after 75 seconds enrollment, shows robustness to noise, motion, and replay attacks.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 0:28The evolution of earphones
- 0:51The lack of knowledge
- 1:13Existing approaches
- 1:46Benefits
- 2:38Unique Information
- 2:55Model
- 3:32Authentication Pipeline
- 4:37Experiments
- 5:22Conclusion
- 5:49Outro
related talks
06mVocalLock: Sensing Vocal Tract for Passphrase-Independent User Authentication Leveraging ...
33mYour voice confirms my identity
39m Put Words In My Mouth
49mChipmunk or Pepe? Using Acoustical Analysis to Detect Voice-Channel Fraud
36mFabricated Realities: AI Personas and Simulacra and the Threats They Pose
20mSpeechGuard: Recoverable and Customizable Speech Privacy Protection...
24mCommanderSong: A Systematic Approach for Practical Adversarial Voice Recognition
35mThe UX Of Advanced Voice Biometrics For Authentication
