Undergraduate students from our university's Department AI, Big Data & Management presented research papers at INTERSPEECH 2025, a premier international conference on artificial intelligence and speech recognition, held in Rotterdam, Netherlands from August 17 to 21. The paper presented, titled “Mimic Blocker: Self-Supervised Adversarial Training for Voice Conversion Defense with Pretrained Feature Extractors,” was researched by undergraduate students Yoo Kwang Yeol, Lee Jun Hyuk, Kim Seo Ryeong, and Lee Ji Min, under the guidance of Professor Lee Jae Hyuk.

This research proposes a defense technique against one-shot voice conversion (a technology enabling voice conversion using a single sample), which has recently emerged as a social problem due to its potential misuse in voice phishing and voice fraud. The Mimic Blocker proposed by the research team focuses on reducing the potential for abuse by disrupting the voice conversion model's ability to properly learn the speaker's unique vocal characteristics while maintaining voice quality. It received high praise, particularly for demonstrating consistent defensive effectiveness across various voice conversion models, not just a single model.

Professor Lee Jae Hyuk stated, “This research, led by undergraduate students, holds significant meaning as it presents a technical solution addressing real-world security threats,” adding, “Our department will continue fostering a student-led research culture and strive to develop AI technologies directly applicable to real-world problems.”