ISMIR 2024
San Francisco
Yuhang
Wu (Johann) · 吴昱杭
I'm interested in music technology.
Now / Study
MSc student in Sound & Music Computing at Universitat Pompeu Fabra (UPF), Barcelona.
Now / Work
R&D intern at BMAT Music Innovators.
Published at ISMIR 2024 · Findings of ACL 2024 · IEEE ICETCI 2024
// Haikou → Beijing → Barcelona
Haikou · origin
Beijing · study
Barcelona · basecamp
From Haikou, Hainan, now based in Barcelona. During undergrad at BUPT I started out in computer vision — deep-learning semantic segmentation of remote sensing images — which gradually led me to the question I now care about most: can machines genuinely understand music?
At M.A.P I led the benchmark team behind ChatMusician (Findings of ACL 2024), curating 1,000+ music-theory QA pairs later incorporated into the MMLU multimodal subset, and co-authored an ISMIR 2024 study of LLMs' capability of music understanding and generation.
Now an R&D intern at BMAT, actively exploring the intersection of academia and industry.
// Research interests
Music Information Retrieval
LLMs for Music
Audio Watermarking
Generative Audio
Computer Vision
Signal Processing
Findings of ACL 2024
Bangkok
ChatMusician: Understanding and Generating Music Intrinsically with LLM
IEEE ICETCI 2024
Changchun
Research on Fire Detection Based on the YOLOv9 Algorithm
BMAT Music Innovators
R&D Intern
Interning in R&D, actively exploring the intersection of academia and industry.
Chinese Academy of Sciences, Institute of Automation
Research Assistant · Audio Watermarking
Watermarking in TTS / music GenAI systems: reviewed the literature on shortcomings of existing watermarking systems, reproduced a classic baseline and constructed a complete workflow.
Multimodal Art Projection (M.A.P)
Team Leader (Benchmark) → Team Member (LLM Music Evaluation)
Spearheaded construction of the MusicTheoryBenchmark — 1,000+ music theory QA pairs curated to rigorous academic standards, later incorporated into the MMLU multimodal subset. Then evaluated LLMs (GPT-4, Llama, Qwen, Gemma) on music understanding and generation over ~3,000 samples.
Beijing University of Posts and Telecommunications
Research Assistant · Sight-Singing Scoring
Deep-learning automatic scoring for sight-singing, with data-synthesis methods generating unlimited training samples to reduce dependence on fixed datasets.
Chinese Academy of Sciences, Aerospace Information Research Institute
Research Assistant · Computer Vision
Semantic segmentation for remote sensing: reproduced state-of-the-art methods, adapted network architectures, and validated across 10+ open and proprietary datasets.
Education
2025 – 2027 (expected)
Universitat Pompeu Fabra — MSc in Sound & Music Computing
Barcelona. MIR, machine learning for music, generative algorithms, computational musicology.
2021 – 2025
Queen Mary University of London & BUPT — B.Eng. Internet of Things
Beijing · joint programme. First Class Honours · GPA 90/100 · top 8% (rank 15/186)