Your Voice is My Passport

Explore the vulnerabilities of voice fingerprinting authentication in this 36-minute Black Hat conference talk. Delve into recent machine learning advancements enabling the generation of synthetic, high-quality audio using speech recordings. Learn about text-to-speech systems, open-source models like Tacotron and WaveNet, and techniques such as data augmentation and transfer learning. Understand the implications for financial institutions, home automation products, and hi-tech offices. Discover potential mitigation strategies, including multi-factor authentication and CGI detection. Gain insights into the broader context of machine learning in offensive security.

Syllabus

Intro
blackhat Obligatory Sneakers Reference
black hat Overview of Text to Speech (TTS)
black hat Proof of concept
black hat Open Source Text to Speech Models
black hat Model overview
black hat Tacotron v1/v2 and WaveNet
black hat Collecting Data
black hat Data Augmentation: Shifting Pitch
black hat Adding Transfer Learning
black hat Transfer Learning Demo
black hat Putting it all together
black hat Our work in perspective (ML for offense)
black hat Mitigation: MFA
black hat Mitigation: Detect CGA
Conclusions
black hat Final note