| 注册
首页|期刊导航|无线电通信技术|基于深度强化学习的和声自动生成算法

基于深度强化学习的和声自动生成算法

刘至洋 刘维莎 冉黎琼 徐康镭 蒋宇河 李庆 乔少杰

无线电通信技术2024,Vol.50Issue(5):985-992,8.
无线电通信技术2024,Vol.50Issue(5):985-992,8.DOI:10.3969/j.issn.1003-3114.2024.05.017

基于深度强化学习的和声自动生成算法

Harmony Autogeneration Algorithm Based on Deep Reinforcement Learning

刘至洋 1刘维莎 2冉黎琼 1徐康镭 1蒋宇河 1李庆 1乔少杰1

作者信息

  • 1. 成都信息工程大学软件工程学院,四川成都 610225
  • 2. 华盛顿大学应用数学系,华盛顿西雅图 98105
  • 折叠

摘要

Abstract

Machine-automated composition is a cross-field research area that combines artificial intelligence and composition theo-ries.The automated composition algorithm aims to assist users in music creation and helps users reduce workload or provides inspira-tion.Faced with the demand for contemporary music creation,existing automatic composition methods fail to effectively express musical features,and the produced harmonies lack musical structures,failing to meet the diversity of contemporary music creation.To solve the aforementioned problems,a harmonic automatic generation algorithm based on Deep Reinforcement Learning(DRL)and harmonic quantization is proposed,called Automatic Harmony Generation Algorithm Based on DRL(AHG-DRL).AHG-DRL uses a harmonic quantization method to encode music,making the encoded music have more comprehensive musical features.It also uses a harmony gener-ation algorithm based on Inverse Reinforcement Learning(IRL)and DRL to expand the search space of music creation while making the generated music have functionality.Experimental results indicate that the proposed automatic harmony generation algorithm can generate music that meets creative requirements and conforms to composition rules.When compared with other harmony generation algorithms,it can generate more complex and diverse types of harmonies,which approximates to the ground-truth values in the objective evaluation metrics.

关键词

和声量化/深度强化学习/自动作曲/预训练/逆强化学习

Key words

harmony quantization/DRL/automatic composition/pretraining/IRL

分类

信息技术与安全科学

引用本文复制引用

刘至洋,刘维莎,冉黎琼,徐康镭,蒋宇河,李庆,乔少杰..基于深度强化学习的和声自动生成算法[J].无线电通信技术,2024,50(5):985-992,8.

基金项目

四川省科技创新苗子工程项目(MZGC20230076) (MZGC20230076)

国家自然科学基金(62272066) (62272066)

四川省科技计划(2021JDJQ0021,2022YFG0186,2022NSFSC0511,2023YFG0027) (2021JDJQ0021,2022YFG0186,2022NSFSC0511,2023YFG0027)

教育部人文社会科学研究规划基金(22YJAZH088) (22YJAZH088)

宜宾市引进高层次人才项目(2022YG02) (2022YG02)

成都市"揭榜挂帅"科技项目(2022-JB00-00002-GX,2021-JB00-00025-GX) (2022-JB00-00002-GX,2021-JB00-00025-GX)

成都市技术创新研发项目(重点项目)(2024-YF08-00029-GX) (重点项目)

成都市区域科技创新合作项目(2023-YF11-00020-HZ) Sichuan Science and Technology Innovation Seedling Project(MZGC20230076) (2023-YF11-00020-HZ)

National Natural Science Foundation of China(62272066) (62272066)

Sichuan Science and Technology Program(2021JDJQ0021,2022YFG0186,2022NSFSC0511,2023YFG0027) (2021JDJQ0021,2022YFG0186,2022NSFSC0511,2023YFG0027)

Planning Foundation for Hu-manities and Social Sciences of Ministry of Education of China(22YJAZH088) (22YJAZH088)

High-level Talent Introduction Project of Yibin(2022YG02) (2022YG02)

Chengdu"Take the Lead"Science and Technology Project(2022-JB00-00002-GX,2021-JB00-00025-GX) (2022-JB00-00002-GX,2021-JB00-00025-GX)

Chengdu Technological Innovation Research and Development Major Project(2024-YF08-00029-GX) (2024-YF08-00029-GX)

Chengdu Regional Science and Technology Innovation Cooperation Project(2023-YF11-00020-HZ) (2023-YF11-00020-HZ)

无线电通信技术

OA北大核心

1003-3114

访问量0
|
下载量0
段落导航相关论文