Hyperbolic Music Transformer for Structured Music Generation
In the field of music generation, generating structured music is a highly challenging research topic. Music generation methods are currently learned in Euclidean space and usually modeled as a time series without structural properties, but due to the limitations of the time series representation in...
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2023-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/10070602/ |
| _version_ | 1797861315721560064 |
|---|---|
| author | Wenkai Huang Yujia Yu Haizhou Xu Zhiwen Su Yu Wu |
| author_facet | Wenkai Huang Yujia Yu Haizhou Xu Zhiwen Su Yu Wu |
| author_sort | Wenkai Huang |
| collection | DOAJ |
| description | In the field of music generation, generating structured music is a highly challenging research topic. Music generation methods are currently learned in Euclidean space and usually modeled as a time series without structural properties, but due to the limitations of the time series representation in Euclidean space, the hierarchical structure of music is difficult to learn, and the generated music is poorly structured. Therefore, based on hyperbolic theory, this paper proposes a Hyperbolic Music Transformer model, which considers the hierarchy in music and models the structured components of music in hyperbolic space. Meanwhile, in order for the network to have sufficient capacity to learn music data with hierarchical and power regular structure, a hyperbolic attention mechanism is proposed, which is an extension of the attention mechanism in hyperbolic space based on the definition of hyperboloid and Klein model. Subjective and objective experiments show that the model proposed in this paper is able to generate high-quality music with structure. |
| first_indexed | 2024-04-09T22:01:32Z |
| format | Article |
| id | doaj.art-eab925a6bf8f4605bd6af30fd2c4a62e |
| institution | Directory Open Access Journal |
| issn | 2169-3536 |
| language | English |
| last_indexed | 2024-04-09T22:01:32Z |
| publishDate | 2023-01-01 |
| publisher | IEEE |
| record_format | Article |
| series | IEEE Access |
| spelling | doaj.art-eab925a6bf8f4605bd6af30fd2c4a62e2023-03-23T23:00:30ZengIEEEIEEE Access2169-35362023-01-0111268932690510.1109/ACCESS.2023.325738110070602Hyperbolic Music Transformer for Structured Music GenerationWenkai Huang0https://orcid.org/0000-0003-3111-7511Yujia Yu1https://orcid.org/0000-0002-2851-4340Haizhou Xu2https://orcid.org/0000-0001-7288-9732Zhiwen Su3https://orcid.org/0000-0003-3500-5955Yu Wu4https://orcid.org/0000-0002-1804-0495School of Mechanical and Electrical Engineering, Guangzhou University, Guangzhou, ChinaSchool of Mechanical and Electrical Engineering, Guangzhou University, Guangzhou, ChinaSchool of Mechanical and Electrical Engineering, Guangzhou University, Guangzhou, ChinaSchool of Mechanical and Electrical Engineering, Guangzhou University, Guangzhou, ChinaLaboratory Center, Guangzhou University, Guangzhou, ChinaIn the field of music generation, generating structured music is a highly challenging research topic. Music generation methods are currently learned in Euclidean space and usually modeled as a time series without structural properties, but due to the limitations of the time series representation in Euclidean space, the hierarchical structure of music is difficult to learn, and the generated music is poorly structured. Therefore, based on hyperbolic theory, this paper proposes a Hyperbolic Music Transformer model, which considers the hierarchy in music and models the structured components of music in hyperbolic space. Meanwhile, in order for the network to have sufficient capacity to learn music data with hierarchical and power regular structure, a hyperbolic attention mechanism is proposed, which is an extension of the attention mechanism in hyperbolic space based on the definition of hyperboloid and Klein model. Subjective and objective experiments show that the model proposed in this paper is able to generate high-quality music with structure.https://ieeexplore.ieee.org/document/10070602/Structured music generationhyperbolic theoryhyperbolic attentionhyperbolic music transformer |
| spellingShingle | Wenkai Huang Yujia Yu Haizhou Xu Zhiwen Su Yu Wu Hyperbolic Music Transformer for Structured Music Generation IEEE Access Structured music generation hyperbolic theory hyperbolic attention hyperbolic music transformer |
| title | Hyperbolic Music Transformer for Structured Music Generation |
| title_full | Hyperbolic Music Transformer for Structured Music Generation |
| title_fullStr | Hyperbolic Music Transformer for Structured Music Generation |
| title_full_unstemmed | Hyperbolic Music Transformer for Structured Music Generation |
| title_short | Hyperbolic Music Transformer for Structured Music Generation |
| title_sort | hyperbolic music transformer for structured music generation |
| topic | Structured music generation hyperbolic theory hyperbolic attention hyperbolic music transformer |
| url | https://ieeexplore.ieee.org/document/10070602/ |
| work_keys_str_mv | AT wenkaihuang hyperbolicmusictransformerforstructuredmusicgeneration AT yujiayu hyperbolicmusictransformerforstructuredmusicgeneration AT haizhouxu hyperbolicmusictransformerforstructuredmusicgeneration AT zhiwensu hyperbolicmusictransformerforstructuredmusicgeneration AT yuwu hyperbolicmusictransformerforstructuredmusicgeneration |