Peiling
Lu.
Artificial intelligence, music, speech. Teaching machines to compose, sing, and listen.
MuseCoco
Generates symbolic music from natural-language descriptions — "slow, contemplative, 3/4, piano & flute, religious." Two stages: text → musical attributes → notes you can edit in MuseScore. 1.2B-parameter model.
Selected works
-
[02]↗Museformer: Transformer with Fine- and Coarse-Grained Attention for Music GenerationYu, Lu, Wang, Hu, Tan, Ye, Zhang, Qin, LiuNeurIPS 2022 147 cites
-
[03]↗XiaoiceSing: a high-quality & integrated singing voice synthesis systemLu, Wu, Luan, Tan, ZhouInterspeech 2020 first author 140 cites
-
[04]↗TeleMelody: lyric-to-melody generation with a template-based two-stage methodJu, Lu, Tan, Wang, Zhang, Wu, Zhang, Li, Qin, LiuEMNLP 2022 63 cites
-
[05]↗MusicAgent: an AI agent for music understanding & generation with LLMsYu, Song, Lu, He, Tan, Ye, Zhang, BianEMNLP 2023 demo 52 cites
-
[06]↗MeloForm: generating melody with musical form based on expert systems & neural networksLu, Tan, Yu, Qin, Zhao, LiuISMIR 2022 first author 25 cites
-
[07]↗EmoGen: eliminating subjective bias in emotional music generationKang, Lu, Yu, Tan, Ye, Zhang, BianarXiv 2023 13 cites
About the researcher.
A researcher at Microsoft, building deep-learning systems that compose, sing, and structure music. Contributor to the Muzic open-source project at Microsoft Research Asia — home to MuseCoco, Museformer, MusicAgent, XiaoIceSing & more. Currently extending that craft into self-service GenAI and agentic systems.