SpoofCeleb: Speech Deepfake Detection and SASV in the Wild
About this Topic:
This talk presents our presenters team’s work across two papers—Text-to-Speech in the Wild and SpoofCeleb. Current speech deepfake detection systems are mostly trained on datasets containing clean speech recorded in controlled environments, along with synthesized speech generated by TTS models also trained on clean data. However, real-world deployment conditions are diverse and noisy, creating a mismatch. TTS model’s dependence on clean training data has long limited the creation of realistic in-the-wild anti-spoofing datasets.
In this talk, our presenter will describe how they tackle this challenge and built a large-scale in-the-wild dataset for training and evaluating robust speech deepfake detection systems.
About the Presenter:
Jee-weon Jung (M’ 18) received the B.S. and the Ph.D. degrees in computer science & engineering from the University of Seoul, Korea, in 2017 and 2021, respectively,
He is currently a senior research scientist at Apple and was a research scientist at Naver corporation and a Postdoctoral Research Associate at Carnegie Mellon University, USA. He is an organizer of ASVspoof, VoxSRC, SASV, and WildSpoof series. His research focuses on audio speaker recognition, speech deepfake detection, and other speech-related tasks.
Want to learn more about upcoming events & webinars?
Visit the events section of the Signal Processing website to see all upcoming lectures, workshops, webinars, and more.