Put Words In My Mouth
summary
Demonstrates that convincing voice clones can be made from tiny samples — down to five seconds of audio — using tools like Lyrebird and Tacotron (Adobe’s VoCo and Google Duplex also shown). A proof‑of‑concept voice note fools a father and a 25‑year friend; a recorded industry incident let attackers authorize a $243,000 transfer. Attack flow uses spectrogram-based synthesis, virtual audio cables and Skype calls; background noise increases believability. Recommendation: opt out of voice‑only auth and use multi‑factor.
AI-written summary · about
chapters (11)
- 0:00Start
- 3:08Iot Devices as the Infiltration of Things
- 5:15How a Voice Is Made
- 8:11Spectrogram
- 8:32Voice Verification
- 13:06Google Voice
- 13:54Google's Assistant
- 18:21How Does Deep Throat Actually Work
- 20:08Background Noise
- 23:22Live Demo
- 31:43Deep Fakes
related talks
33mYour voice confirms my identity
36mFabricated Realities: AI Personas and Simulacra and the Threats They Pose
39mSynthetic Trust: Exploiting Biases at Scale
47mFighting AI with AI
36mDemocratized Deception: How Threat Actors Are Weaponizing AI to Scale Cybercrime
30mHidden Voice Commands
24mCommanderSong: A Systematic Approach for Practical Adversarial Voice Recognition
45mVoice Security - Get Up to Speed on This Attack Vector