Weixi Zhai

Music Information Retrieval · Generative AI for Music

I work on music information retrieval and generative music, focusing on performance representation and expressive generation. I study relationships between scores and piano performances, including timing, dynamics, and variation across levels of expertise. My work also explores Nanyin ensemble generation, using pipa-centric data and graph representations to model heterophony and ornamentation. Across these projects, I investigate music tokenization, latent-space interpolation, and timbre-based interfaces for creative interaction. I studied music at Quanzhou Normal University (MA, 2023–2026), supervised by Prof. Jianbing Xiahou.

Weixi Zhai

Publications

Google Scholar

Interactive Performance Works

Spiral trajectories and sound-analysis controls in Real-Time Timbral ExplorationsInteractive performance · video documentation

2025

Real-Time Timbral Explorations

A Visual Journey Through MIR Algorithms

An interactive performance system that uses MIR algorithms as a compositional mechanism. Spiral trajectories navigate timbre spaces, while a visual interface reveals algorithmic states and sonic transformations.

Real-time MFCC analysis · k-NN interpolation · Granular synthesis · Archimedean, logarithmic, and Fermat spiral trajectories

Live Performance Systems

ISMIR 2025 livestream cover for the recording containing the RenCon performanceRenCon 2025 · ISMIR livestream recording

2025 · Finalist & System Developer · ISMIR 2025, Daejeon, South Korea

ScorePerLock

RenCon 2025: Expressive Piano Performance Rendering Contest

An expressive piano rendering system that models phrasing, dynamics, and timing from symbolic scores. Its renderings were played on a Yamaha Disklavier in the RenCon 2025 live concert, where listeners rated anonymized performances and tried to identify the human reference.

Symbolic score-to-performance rendering · Learned phrasing, dynamics, and timing