MUUNRiver-Bench: Diagnosing Relation-Dependent Music Retrieval with Multimodal Instructions
研究针对音乐检索中关系依赖性问题，通过MUUNRiver-Bench基准，使用多模态指令来定义相关性，揭示了不同模型在处理特定任务时的偏好差异。
0 citationsRead paper
Selected work
Representative Papers
TUTTI: Toward generalizable audio-to-score transcription via fully synthesized data
为解决高质量音频到乐谱转录数据稀缺问题，研究通过完全合成的数据和Transformer模型预训练方法，提高跨乐器转录性能。
0 citationsRead paper
Dissonance Spectrum explicitly models perceptual frequency interactions for better music understanding
该研究通过引入Dissonance Spectrum来显式建模同时频率成分间的关系，以改进音乐理解，并在多项音乐理论测试中表现出色。
0 citationsRead paper
UniVerse: Benchmarking and Enhancing LALMs on Culturally Inclusive Low-Resource Music Understanding
为解决大型音频-语言模型在文化多样性音乐理解上的不足，研究通过构建包含5042个问答对的基准测试集UniVerseBench及自动化生成的训练数据集UniVerseSet，采用不平衡学习策略改进模型性能。
0 citationsRead paper
Where Does AI Innovation Go? Measuring Research Attention Imbalance in AI Music
This study addresses the rapid yet uneven growth of artificial intelligence research in music, where attention across tasks lacks systematic measurement. To characterize this imbalance, the authors construct a joint taxonomy encompassing 12 application domains and 11 technical methodologies, analyzing 6,839 publications from 2015 to 2026. They propose a novel four-dimensional profile of research attention—comprising technical investment, method allocation, methodological diversity, and adoption lag of cutting-edge techniques—and apply bibliometric analysis combined with time-lag modeling. The findings reveal that generative tasks adopt state-of-the-art methods rapidly (average lag: 0.33 years), whereas applications in education, health, and governance exhibit substantial delays (4.33 and 5.00 years, respectively), highlighting significant disparities in research resource allocation and filling a critical gap in assessing equity within AI-driven music research.
0 citationsRead paper
Recent publications
Latest Papers
MUUNRiver-Bench: Diagnosing Relation-Dependent Music Retrieval with Multimodal Instructions
研究针对音乐检索中关系依赖性问题，通过MUUNRiver-Bench基准，使用多模态指令来定义相关性，揭示了不同模型在处理特定任务时的偏好差异。
0 citationsRead paper
TUTTI: Toward generalizable audio-to-score transcription via fully synthesized data
为解决高质量音频到乐谱转录数据稀缺问题，研究通过完全合成的数据和Transformer模型预训练方法，提高跨乐器转录性能。
0 citationsRead paper
Dissonance Spectrum explicitly models perceptual frequency interactions for better music understanding
该研究通过引入Dissonance Spectrum来显式建模同时频率成分间的关系，以改进音乐理解，并在多项音乐理论测试中表现出色。
0 citationsRead paper
UniVerse: Benchmarking and Enhancing LALMs on Culturally Inclusive Low-Resource Music Understanding
为解决大型音频-语言模型在文化多样性音乐理解上的不足，研究通过构建包含5042个问答对的基准测试集UniVerseBench及自动化生成的训练数据集UniVerseSet，采用不平衡学习策略改进模型性能。
0 citationsRead paper
Where Does AI Innovation Go? Measuring Research Attention Imbalance in AI Music
This study addresses the rapid yet uneven growth of artificial intelligence research in music, where attention across tasks lacks systematic measurement. To characterize this imbalance, the authors construct a joint taxonomy encompassing 12 application domains and 11 technical methodologies, analyzing 6,839 publications from 2015 to 2026. They propose a novel four-dimensional profile of research attention—comprising technical investment, method allocation, methodological diversity, and adoption lag of cutting-edge techniques—and apply bibliometric analysis combined with time-lag modeling. The findings reveal that generative tasks adopt state-of-the-art methods rapidly (average lag: 0.33 years), whereas applications in education, health, and governance exhibit substantial delays (4.33 and 5.00 years, respectively), highlighting significant disparities in research resource allocation and filling a critical gap in assessing equity within AI-driven music research.
0 citationsRead paper