Li Wang

Li Wang

Ph.D. Candidate, CUHK-Shenzhen

I am a Ph.D. candidate in Computer Science at The Chinese University of Hong Kong, Shenzhen (CUHK-Shenzhen), advised by Professor Zhizheng Wu and Professor Haizhou Li.

My research focuses on trustworthy speech and audio intelligence, especially audio deepfake detection, spoken misinformation detection, speaker verification security, and audio language models. I am interested in building reliable AI systems that can detect, explain, and defend against synthetic or adversarial audio in realistic scenarios.

Before joining CUHK-Shenzhen, I received my master’s degree from Peking University (北京大学), where I worked on spoken keyword spotting and speaker verification with Professor Yuexian Zou (邹月娴). I received my bachelor’s degree from University of Electronic Science and Technology of China (电子科技大学), majoring in Software Engineering.

My work has appeared in IEEE/ACM TASLP, ICLR, IEEE SLT, ICASSP, and Interspeech. My recent first-author paper GenTraceBench was accepted by IEEE ISCSLP 2026, and my first-author manuscript RealComm was submitted to IEEE/ACM TASLP. I have also worked on industry-facing speech and audio projects with Huawei, ByteDance, Tencent AI Lab, Xiaomi, Shenzhen Research Institute of Big Data, and International Digital Economy Academy.

[Google Scholar] [GitHub] [ORCID] [CV (English)] [CV (Chinese)]

🔥 News

  • 2026.09: My first-author paper GenTraceBench was accepted by IEEE ISCSLP 2026, and my first-author manuscript RealComm was submitted to IEEE/ACM TASLP. [Code]
  • 2026.09: I contributed to the audio deepfake detection capability integrated into Huawei HarmonyOS 7.

📝 Publications

7 peer-reviewed papers, 1 accepted paper, 1 submitted manuscript, and 3 technical reports/preprints.

  1. GenTraceBench: A Benchmark for Tracing Audio Deepfakes Across Pre- and Post-training Stages
    Li Wang, Kunyu Feng, Wan Lin, Dekun Chen, Qinke Ni, Xueyao Zhang, Lei Wang, Jie Shi, Haizhou Li, Zhizheng Wu.
    IEEE ISCSLP 2026 (Accepted). [Paper]

  2. RealComm: Benchmarking and Adapting Audio Deepfake Detection over Real Communication Channels
    Li Wang†, Jindong Wang†, Wan Lin†, Kunyu Feng†, Lei Wang, Rizhao Cai, Ce Fang, Jinzhe Xue, Jie Shi, Haizhou Li, Zhizheng Wu. († Equal contribution.)
    Submitted to IEEE/ACM Transactions on Audio, Speech, and Language Processing (TASLP). [Project] [Paper] [Code]

  3. Teffic-Audio: Tell Fact from Fiction
    Wan Lin, Li Wang, Jindong Wang, Kunyu Feng, Zhizheng Wu.
    Technical Report, 2026. [Paper] [Project]

  4. VoxSafeBench: Not Just What Is Said, but Who, How, and Where
    Yuxiang Wang, Hongyu Liu, Yijiang Xu, Qinke Ni, Li Wang, Wan Lin, Kunyu Feng, Dekun Chen, Xu Tan, Lei Wang, Jie Shi, Zhizheng Wu.
    Preprint, 2026. [Paper] [Project] [Code]

  5. DFALLM: Achieving Generalizable Multitask Deepfake Detection by Optimizing Audio LLM Components
    Yupei Li†, Li Wang†, Yuxiang Wang, Lei Wang, Rizhao Cai, Jie Shi, Björn W. Schuller, Zhizheng Wu. († Equal contribution.)
    Technical Report, 2025. [Paper]

  6. SpeechJudge: Towards Human-Level Judgment for Speech Naturalness
    Xueyao Zhang, Chaoren Wang, Huan Liao, Ziniu Li, Yuancheng Wang, Li Wang, Dongya Jia, Yuanzhe Chen, Xiulin Li, Zhuo Chen, Zhizheng Wu.
    ICLR 2026. [Paper]

  7. Over-the-Air Adversarial Attacks and Detection for Automatic Speaker Verification
    Li Wang, Xiao Lei, Haorui He, Lei Wang, Jie Shi, Zhizheng Wu.
    IEEE/ACM TASLP 2026. [Paper]

  8. SpMis: An Investigation of Synthetic Spoken Misinformation Detection
    Peizhuo Liu†, Li Wang†, Renqiang He†, Haorui He, Lei Wang, Huadi Zheng, Jie Shi, Tong Xiao, Zhizheng Wu. († Equal contribution.)
    IEEE SLT 2024 — Best Paper Finalist, Top 2.5%. [Paper]

  9. AdvSV: An Over-the-Air Adversarial Attack Dataset for Speaker Verification
    Li Wang, Jiaqi Li, Yuhao Luo, Jiahao Zheng, Lei Wang, Hao Li, Ke Xu, Chengfang Fang, Jie Shi, Zhizheng Wu.
    IEEE ICASSP 2024. [Paper]

  10. An Initial Investigation of Neural Replay Simulator for Over-the-Air Adversarial Perturbations to Automatic Speaker Verification
    Jiaqi Li, Li Wang, Liumeng Xue, Lei Wang, Zhizheng Wu.
    IEEE ICASSP 2024. [Paper]

  11. Learning Decoupling Features Through Orthogonality Regularization
    Li Wang, Rongzhi Gu, Weiji Zhuang, Peng Gao, Yujun Wang, Yuexian Zou.
    IEEE ICASSP 2022. [Paper]

  12. Text Anchor Based Metric Learning for Small-Footprint Keyword Spotting
    Li Wang, Rongzhi Gu, Nuo Chen, Yuexian Zou.
    Interspeech 2021. [Paper]

🏆 Honors and Awards

  • IEEE SLT 2024 Best Paper Finalist, Top 2.5%, for SpMis: An Investigation of Synthetic Spoken Misinformation Detection, 2024.
  • Best Poster Award, Symposium on Speaker Recognition Research and Applications, 2024.
  • Duan Yongping Travel Scholarship, 2023.
  • Runner-up, Peking University 4x100m relay, 2021.
  • National Second Prize, National University Student Information Security Competition, 2018.
  • Undergraduate People’s First-Class Scholarship, University of Electronic Science and Technology of China, 2015, 2016, 2017.
  • Outstanding Volunteer, the 9th National Paralympic Games of China, 2015.

📖 Education

💼 Research and Work Experience

  • 2022.11 - 2023.08, Research Assistant, Shenzhen Research Institute of Big Data (SRIBD), Shenzhen.
    Led a Huawei-collaboration project on detecting physical attacks against speaker verification systems; built a large-scale dataset with 1.1M physical-attack samples and developed a robust defense route for real-world acoustic conditions.
  • 2022.07 - 2022.10, Financial Machine Learning Engineer, International Digital Economy Academy (IDEA), Shenzhen.
    Designed financial factors and deep learning models for stock-return prediction and financial time-series modeling.

💻 Internships

  • 2021.07 - 2022.03, ByteDance AI Lab, Shenzhen.
    Worked on audio fingerprinting and music segmentation; proposed a temporal localization algorithm that improved accuracy from 95.80% to 97.90%.
  • 2021.04 - 2021.06, Huawei MTI, Audio Engineering Department, Shenzhen.
    Worked on music source separation; reproduced D3Net on MUSDB and separated vocals, drums, bass, and other tracks.
  • 2020.05 - 2020.11, Tencent AI Lab, Speech Processing Group, Beijing.
    Worked on speech separation and speaker verification; evaluated GALR-based separation for speaker verification and organized speaker metadata for 2M audio samples.

🚀 Selected Projects

  • Huawei HarmonyOS 7 Audio Deepfake Detection, 2025.01 - Present.
    Contributed to the research, development, and product deployment of the audio deepfake detection capability integrated into HarmonyOS 7. Studied robust detection across diverse speech synthesis methods and real communication channels.
  • Research on AI Voice Wake-up Technology, 2021.01 - 2022.01.
    Project leader in a collaboration with Xiaomi. Improved network design and data augmentation, reducing false rejection rate from 20% to 2% and achieving 99.7% accuracy. The work led to a first-author ICASSP 2022 paper.
  • AdvSV and AdvSV 2.0: Adversarial Speaker Verification Security, 2022 - 2026.
    Built benchmarks for over-the-line and over-the-air adversarial attacks against automatic speaker verification systems, including neural replay simulation and one-class contrastive detection.
  • SpMis: Spoken Misinformation Detection, 2024.
    Built a large-scale synthetic spoken misinformation dataset and detection benchmark covering 1,000+ speakers, multiple topics, and state-of-the-art TTS systems.

📜 Patents and Software Copyright

Patents

  • Audio Deepfake Detection Method, Apparatus, Computer Device, and Storage Medium. Inventors: Li Wang, Zhizheng Wu. Chinese invention patent ZL 2024 1 1863864.6; grant publication CN 119694319 B; granted November 11, 2025.
  • Adversarial Audio Sample Detection Method and System Based on Pretrained Models. Inventors: Li Wang, Zhizheng Wu. Chinese invention patent application 202410942156.5; filed and accepted July 15, 2024.
  • Open-Set Audio Deepfake Algorithm Attribution Method. Inventors: Li Wang, Zhizheng Wu. Chinese invention patent application 202510885063.8; filed and accepted June 30, 2025.
  • Audio Deepfake Detection Method Based on Parameter Discriminative Sensitivity and Terminal Device. Inventors: Kunyu Feng, Li Wang, Wan Lin, Zhizheng Wu. Chinese invention patent application 202610887808.9; filed and accepted June 18, 2026.
  • Audio Deepfake Detection Software V1.0 (音频伪造检测软件 V1.0). Copyright owner: Shenzhen Research Institute of Big Data. Registration No. 2026SR0060775; Certificate No. 软著登字第17275056号; registered January 9, 2026.

🧑‍🏫 Services

🛠 Skills

  • Programming: Python, Java, C++.
  • Deep learning: PyTorch, speech/audio model training, large-scale data processing.
  • Languages: English academic writing and peer review.
  • Other: Erhu Grade 10; track and field.

👏 Template of This Page

Thanks to Yi Ren (任意) for his open source contribution, template link AcadHomepage .