Profile
I am currently a Ph.D. student at the Department of Electronic Engineering of Tsinghua University, supervised by Prof. Pingyi Fan (WISTLAB) and co-advised by Prof. Jia Liu and Prof. Wei-Qiang Zhang (SATLAB). Prior to that, I received my Bachelor's degree in Electronic Engineering from Tsinghua University in 2022.
My research focuses on universal representation of multimodal industrial signals (mostly audio and vibration), and downstream anomaly detection. We want to perceive the world through hearing. Our goal is not just publications, but more importantly, to see technology truly put into practice and to contribute to a better world.
I expect to graduate in June, 2027. I am seeking full-time job in industries, particularly in speech and audio foundation models. Feel free to reach out to me for small talks if you are interested.
Education
Ph.D. student in Electronic Engineering. Expected to graduate in June, 2027.
Supervisor: Prof. Pingyi Fan. Co-advised by Prof. Jia Liu and Prof. Wei-Qiang Zhang.
B.S. in Electronic Engineering
Selected Publications
[IEEE TII] FISHER: A Foundation Model for Multimodal Industrial Signal Comprehensive Representation
FISHER is the first Foundation for multi-modal Industrial Signal compreHEnsive Representation. It novelly introduces sub-band modeling for multi-sampling-rate signals and builds the overall representation by concatenating sub-band representations, just like assembling building blocks. FISHER has delivered outstanding performance in both experimental setups and real machines, and it will soon be deployed in real-world applications.
[ICASSP] Adaptive Prototype Learning for Anomalous Sound Detection with Partially Known Attributes
This paper uses multiple learnable sub-centers for coarse labels and synthetic oversampling to better handle domain shifts, producing more discriminative normal-sound prototypes. Evaluated on DCASE 2024 Task 2, it achieved a 65.01% overall score and surpassed the challenge-winning system on the evaluation set.
[GLOBECOM] CoopASD: Cooperative Machine Anomalous Sound Detection with Privacy Concerns
CoopASD introduces a federated-learning framework that lets factories jointly train machine fault-detection models without sharing sensitive audio data. It combines a pretrained audio transformer with local KNN-based detection and tailored training strategies for highly heterogeneous factory data.
[INTERSPEECH] AnoPatch: Towards Better Consistency in Machine Anomalous Sound Detection
AnoPatch is a groundbreaking work for the field of anomalous sound detection. It fine-tunes a pre-trained model using operating-condition information, and for the first time substantially outperforms traditional non-pretrained models.
[ICASSP] Unsupervised Anomaly Detection and Localization of Machine Audio: A GAN-Based Approach
AEGAN-AD is an unsupervised GAN-based method for detecting anomalous machine sounds from normal-only training data. It combines spectrogram reconstruction with discriminator-derived semantic features to improve robustness under changing operating conditions.
Challenges
- DCASE 2024 Task 2: machine anomalous sound detection — first- and second-place systems.
- ICME 2024 ASC Grand Challenge: acoustic scene classification in urban cities — second- and third-place systems.
- DCASE 2023 Task 2: machine anomalous sound detection — second- and third-place systems.
Services
- Conference reviewer: ICASSP, Interspeech, SLT, NCMMSC.
- Journal reviewer: TASLP, TCCN.
Internships
Core member. Spearheaded the industrialization of R&D AI signal processing. · Beijing, China
Large language models for recommendation · Shenzhen, China
Target-speaker automatic speech recognition · Beijing, China
Model survey and tutorials for PaddleHub · Beijing, China
Awards
- IIEP Research Scholarship, IIEP, March 2026.
- DCASE 2024 Challenge Judges' Award, October 2024.