Audio LLMs that not only detect speech edits but locate where they happen.
Research dossier · 2026
I study how to tell
I study how to tell
synthetic from real.
武汉大学网络空间安全学院博士候选人,研究方向为 多模态深度伪造检测(人脸与语音)、 对抗攻防与声纹隐私保护、音视频取证。 Ph.D. candidate at Wuhan University, researching multimedia security — deepfake detection, adversarial robustness, and audio-visual forensics.
Multimedia SecuritySpeech Deepfake Detection 语音深度伪造检测
Detecting AI-generated and spoofed speech across real-world channels — real-time communication, speaker-specific fingerprints, editing localization, and generalisation gaps.
检测真实信道中的 AI 合成与伪造语音——实时通信、说话人特异性指纹、 语音编辑定位与泛化鸿沟。
Adversarial & Privacy Protection 对抗与隐私保护
Adversarial examples and defenses for speaker recognition and voice privacy — low-cost, universal, and real-time.
面向声纹识别与语音隐私的对抗攻防——低开销、通用、实时。
Multimodal Research 多模态研究
Multi-modal learning across audio and video — lip-sync for talking heads, and audio-visual speech recognition under real-world degradation.
音视频跨模态研究——口型同步数字人合成,以及真实世界退化下的音视频语音识别。
Audio-Visual Forensics 音视频取证
Cross-modal forensics for synthetic faces — wavelet-domain cues and speaking-behaviour facial landmarks expose generated content.
面向合成人脸的跨模态取证——小波域线索与说话行为面部关键点揭露伪造内容。
/01 News & Highlights 动态
News · 新闻
2026.07MM 2026 接收:Time-Frequency Consistency Learning for Robust Speech Deepfake Detection (Poster)
2026.06IEEE TIFS 接收:UnVC: Protecting Your Voiceprint by Generative Adversarial Speech (co-first author)
2026.06Interspeech 2026 接收:Exploring the Scale and Diversity of Speech Anti-spoofing Datasets (Poster)
2026.05IJCAI 2026 接收:Profiling the Voice: Speaker-Specific Phoneme Fingerprinting for Speech Deepfake Detection (Poster)
2026.04ACL 2026 接收:RTCFake: Speech Deepfake Detection in Real-Time Communication (Poster)
2026.03ICME 2026 接收:PASE: Phoneme-Aware Speech Encoder to Improve Lip Sync Accuracy for Talking Head Synthesis (first author, Spotlight)
2026.02CVPR 2026 接收:When AVSR Meets Video Conferencing: Dataset, Degradation, and the Hidden Mechanism Behind Performance Collapse (first author, Poster)
2025.05IEEE TIFS 接收:APFT: Adaptive Phoneme Filter Template to Generate Anti-Compression Speech Adversarial Example in Real-Time (first author)
2025.03ICASSP 2025 接收:FreqSense: Universal and Low-Latency Adversarial Example Detection for Speaker Recognition (first author, Poster)
Honors & Awards · 荣誉
2025国家奖学金 National Scholarship
2025雷军奖学金 Lei Jun Scholarship
2025武汉大学优秀研究生 Outstanding Graduate Student
Experience · 经历
2026.06 – now结构化数据融合大模型研究实习生 · 风控 · 筋斗云人才计划(27届)
2024 – now博士研究生 · 武汉大学网络空间安全学院
ReviewerIEEE TIFS · IEEE/ACM TASLP · ICME 2026 · ICML 2026
/02 Publications 论文
Grouped by research direction. Underlined name marks me as author. Each paper is tagged with an evidence index — you can look up any of them on Google Scholar.
Speech Deepfake Detection 语音深度伪造检测
7 papersHow diverse must a spoofing dataset be before a detector generalises?
Replay-aware data for the physical-world deepfake threat.
Benchmarking how far today's detectors travel from lab to the real world.
Enforcing time-frequency consistency as a robust cue that survives real-world compression.
Speaker-specific phoneme fingerprints make spoofing a voice much harder than matching it.
Real-time communication is the hardest channel — RTCFake brings deepfake detection into it.
Adversarial & Privacy Protection 对抗与隐私保护
4 papersGenerative adversarial speech as a cloak for your voiceprint.
Catching adversarial examples via interpretable frequency-domain cues.
Adversarial examples that survive audio compression in real time.
A single defense that generalises across speaker-recognition attacks.
Multimodal Research 多模态研究
2 papersPhoneme-aware encoding makes lips and speech agree.
Why audio-visual speech recognition collapses on real video calls.
Audio-Visual Forensics 音视频取证
2 papers
EV-14
一作中文期刊——基于说话行为面部关键点的鲁棒伪造人脸检测。
Wavelet-domain cues expose synthetic talking faces.
/03 Education 教育经历
2015 – 2018
高中 · 福建省永定第一中学
High school, Yongding No.1 Middle School, Fujian
2018 – 2022
本科 · 武汉大学网络空间安全学院 (B.Eng.)
B.Eng., School of Cyber Science and Engineering, WHU
2022 – 2024
硕士 · 武汉大学网络空间安全学院 (M.Eng.)
M.Eng., School of Cyber Science and Engineering, WHU
2024 – 今
博士 · 武汉大学网络空间安全学院 (Ph.D.)
Ph.D., School of Cyber Science and Engineering, WHU
/04 Notes from the Lab 实验手记
A blog about deepfake detection, adversarial robustness, and the craft of working in multimedia security — written for anyone walking the same road. 这里记录研究中的思考与实验笔记。
Why deepfake detection is really a data problem
Every detector I have worked on eventually bumps into the same wall: the dataset. Notes from the lab on why we should...