基於GPT-2和近端策略優化的符號音樂生成

以作者查詢圖書館館藏

、以作者查詢臺灣博碩士

、以作者查詢全國書目

、勘誤回報

、線上人數：45

、訪客IP：3.141.40.192

姓名

張翔威(Hsiang-Wei Chang) 查詢紙本館藏

畢業系所

資訊工程學系

論文名稱

基於GPT-2和近端策略優化的符號音樂生成
(Symbolic Music Generation Using GPT-2 and Proximal Policy Optimization)

相關論文

★ 基於edX線上討論板社交關係之分組機制	★ 利用Kinect建置3D視覺化之Facebook互動系統
★ 利用 Kinect建置智慧型教室之評量系統	★ 基於行動裝置應用之智慧型都會區路徑規劃機制
★ 基於分析關鍵動量相關性之動態紋理轉換	★ 基於保護影像中直線結構的細縫裁減系統
★ 建基於開放式網路社群學習環境之社群推薦機制	★ 英語作為外語的互動式情境學習環境之系統設計
★ 基於膚色保存之情感色彩轉換機制	★ 一個用於虛擬鍵盤之手勢識別框架
★ 分數冪次型灰色生成預測模型誤差分析暨電腦工具箱之研發	★ 使用慣性傳感器構建即時人體骨架動作
★ 基於多台攝影機即時三維建模	★ 基於互補度與社群網路分析於基因演算法之分組機制
★ 即時手部追蹤之虛擬樂器演奏系統	★ 基於類神經網路之即時虛擬樂器演奏系統

檔案

[Endnote RIS 格式]

[Bibtex 格式]

[相關文章]

[文章引用]

[完整記錄]

[館藏目錄]

至系統瀏覽論文 (2027-1-8以後開放)

摘要(中)

本研究旨在探索和實現基於GPT-2和近端策略優化（PPO）強化學習的
符號音樂生成方法。音樂生成在人工智慧與機器學習領域中一直是重要
的研究課題之一，目的是通過算法自動生成具有藝術價值和情感表達的
音樂作品。本研究通過探討GPT-2在生成旋律、伴奏及和弦方面的應用潛
力，並引入PPO強化學習算法，提升生成音樂的質量和多樣性。研究結果
表明，結合GPT-2和PPO的符號音樂生成方法能夠有效提升音樂的創造性
和多樣性，為音樂創作和應用帶來新的可能性。本研究不僅在理論上具有
重要意義，還在實際應用中具有廣泛的前景。

摘要(英)

This study aims to explore and implement a symbolic music generation method based
on GPT-2 and Proximal Policy Optimization (PPO) reinforcement learning. Music gen
eration has long been an important research topic in the fields of artificial intelligence
and machine learning, aiming to automatically generate music works with artistic value
and emotional expression through algorithms. This research explores the application
potential of GPT-2 in generating melodies, accompaniments, and chords, and introduces
the PPO reinforcement learning algorithm to improve the quality and diversity of gen
erated music. The results show that the symbolic music generation method combining
GPT-2 and PPO can effectively enhance the creativity and diversity of music, bringing
new possibilities to music creation and application. This research is of great theoretical
significance and has broad prospects for practical application.

關鍵字(中)

★ 音樂生成
★ 近端策略優化
★ 強化學習
★ 符號音樂生成

關鍵字(英)

★ GPT-2

論文目次

摘要 i
Abstract ii
圖目錄 v
表目錄 vi
1 Introduction 1
1.1 ResearchBackground . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 ResearchMotivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 ResearchObjectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4 ResearchContributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2 RelatedWork 9
3 Method 16
3.1 SymbolicMusicGenerationMethod . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.2 GPT-2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.3 ReinforcementLearning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.4 RewardFunctions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.4.1 RewardFunctionsforMelody . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.4.2 RewardFunctionsforAccompaniment . . . . . . . . . . . . . . . . . . . . . 24
3.4.3 RewardFunctionsforChord . . . . . . . . . . . . . . . . . . . . . . . . . . 25
4 ExperimentDesignandResult 27
4.1 Dataset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
4.1.1 MelodyDataset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
4.1.2 AccompanimentDataset . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
4.1.3 ChordDataset. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.2 ExperimentDesignandResult . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
4.2.1 MelodyGeneration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.2.2 AccompanimentGeneration . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.2.3 ChordGeneration. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
5 Conclusion 48
Reference 51

參考文獻

[1] P. Shang, S. Ni, and L. Zhou, “A probabilistic and random method for the generation of bai
nationality music fragments,” in 2021 IEEE 4th International Conference on Multimedia Infor
mation Processing and Retrieval (MIPR), 2021, pp. 303–307.
[2] L. Mou, Y. Sun, Y. Tian, Y. Sun, Y. Liu, Z. Zhang, R. He, J. Li, J. Li, Z. Li, F. Gao, Y. Shi, and
R. Jain, “Memomusic 3.0: Considering context at music recommendation and combining music
theory at music generation,” in 2023 IEEE International Conference on Multimedia and Expo
Workshops (ICMEW), 2023, pp. 296–301.
[3] W.Wang,X.Li, C.Jin, D. Lu, Q. Zhou, and Y. Tie, “Cps: Full-song and style-conditioned music
generation with linear transformer,” in 2022 IEEE International Conference on Multimedia and
Expo Workshops (ICMEW), 2022, pp. 1–6.
[4] A. Remesh, A. P. K, and M. S. Sinith, “Symbolic domain music generation system based on lstm
architecture,” in 2022 Second International Conference on Next Generation Intelligent Systems
(ICNGIS), 2022, pp. 1–4.
[5] D. Gangal and Y. Kadam, “Unleashing the melodic potential: Music generation with char rnns,”
in 2023 2nd International Conference on Futuristic Technologies (INCOFT), 2023, pp. 1–6.
[6] S. S. Patil, S. H. Patil, A. M. Pawar, R. Shandilya, A. K. Kadam, R. B. Jadhav, and M. S. Bewoor,
“Music generation using rnn-lstm with gru,” in 2023 International Conference on Integration of
Computational Intelligent System (ICICIS), 2023, pp. 1–5.
[7] M. Singhal, B. Saxena, A. P. Singh, and A. Baranwal, “Study of the effectiveness of generative
adversarial networks towards music generation,” in 2023 Second International Conference on
Informatics (ICI), 2023, pp. 1–5.
[8] C.-F. Huang and C.-Y. Huang, “Emotion-based ai music generation system with cvae-gan,” in
2020 IEEE Eurasia Conference on IOT, Communication and Engineering (ECICE), 2020, pp.
220–222.
[9] S. Sajad, S. Dharshika, and M. Meleet, “Music generation for novices using recurrent neural
network (rnn),” in 2021 International Conference on Innovative Computing, Intelligent Commu
nication and Smart Electrical Systems (ICSES), 2021, pp. 1–6.
[10] J. Wang and C. Li, “Chinese style pop music generation based on recurrent neural network,” in
2022 IEEE 5th Advanced Information Management, Communicates, Electronic and Automation
Control Conference (IMCEC), vol. 5, 2022, pp. 513–516.
[11] L. Yi, H. Hu, J. Zhao, and G. Xia, “Accomontage2: A complete harmonization and accompani
ment arrangement system,” 2022.
[12] Z. Wang, Y. Zhang, Y. Zhang, J. Jiang, R. Yang, J. Zhao, and G. Xia, “Pianotree vae: Structured
representation learning for polyphonic music,” 2020.
[13] Y. Zhao, X. Liu, and T. Su, “Piano accompaniment features and performance processing based
on music feature matching algorithm,” in 2021 IEEE International Conference on Advances in
Electrical Engineering and Computer Applications (AEECA), 2021, pp. 525–529.
[14] H. Niu, “Accompaniment generation based on deep learning and genetic algorithm,” in 2023
IEEE International Conference on Control, Electronics and Computer Technology (ICCECT),
2023, pp. 58–65.
[15] H. Liu, “Improvisational dance piano accompaniment system based on bp neural network,” in
2022 International Conference on Computers and Artificial Intelligence Technologies (CAIT),
2022, pp. 21–25.
[16] Q. Wang, S. Zhang, and L. Zhou, “Emotion-guided music accompaniment generation based on
variational autoencoder,” in 2023 International Joint Conference on Neural Networks (IJCNN),
2023, pp. 1–8.
[17] B. Banar and S. Colton, “Autoregressive self-evaluation: A case study of music generation using
large language models,” in 2023 IEEE Conference on Artificial Intelligence (CAI), 2023, pp.
264–265.
[18] N. Imasato, K. Miyazawa, C. Duncan, and T. Nagai, “Using a language model to generate music
in its symbolic domain while controlling its perceived emotion,” IEEE Access, vol. 11, pp.
52412–52428, 2023.
[19] M. R. Bjare, S. Lattner, and G. Widmer, “Exploring Sampling Techniques for Generating
Melodies With a Transformer Language Model,” in Proceedings of the 24th International
Society for Music Information Retrieval Conference. ISMIR, Dec. 2023, pp. 810–816.
[Online]. Available: https://doi.org/10.5281/zenodo.10265411
[20] J. Liu, Y. Dong, Z. Cheng, X. Zhang, X. Li, F. Yu, and M. Sun, “Symphony Generation with
Permutation Invariant Language Model,” in Proceedings of the 23rd International Society for
Music Information Retrieval Conference. ISMIR, Nov. 2022, pp. 551–558. [Online]. Available:
https://doi.org/10.5281/zenodo.7316722
[21] Y. Guo, Y. Liu, T. Zhou, L. Xu, and Q. Zhang, “An automatic music generation and evaluation
method based on transfer learning,” PLOS ONE, vol. 18, no. 5, pp. 1–21, 05 2023. [Online].
Available: https://doi.org/10.1371/journal.pone.0283103
[22] A. Radford, J. Wu, R. Child, D. Luan, D. Amodei, and I. Sutskever, “Language models are
unsupervised multitask learners,” 2019.
[23] A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. Kaiser, and
I. Polosukhin, “Attention is all you need,” 2023.
[24] R. S. Sutton and A. G. Barto, Reinforcement Learning: An Introduction. MIT Press, 2018.
[25] J. Schulman, F. Wolski, P. Dhariwal, A. Radford, and O. Klimov, “Proximal policy optimization
algorithms,” 2017.
[26] L. von Werra, Y. Belkada, L. Tunstall, E. Beeching, T. Thrush, N. Lambert, and S. Huang, “Trl:
Transformer reinforcement learning,” https://github.com/huggingface/trl, 2020.
[27] V. Mnih, K. Kavukcuoglu, D. Silver, A. A. Rusu, J. Veness, M. G. Bellemare, A. Graves,
M. Riedmiller, A. K. Fidjeland, G. Ostrovski, S. Petersen, C. Beattie, A. Sadik, I. Antonoglou,
H. King, D. Kumaran, D. Wierstra, S. Legg, and D. Hassabis, “Human-level control through
deep reinforcement learning,” Nature, vol. 518, no. 7540, pp. 529–533, 2015.
[28] T. P. Lillicrap, J. J. Hunt, A. Pritzel, N. Heess, T. Erez, Y. Tassa, D. Silver, and D. Wierstra,
“Continuous control with deep reinforcement learning,” 2019.
[29] J.Schulman,S.Levine, P.Moritz, M.I.Jordan, andP.Abbeel, “Trustregionpolicyoptimization,”
2017.
[30] C. Finn, P. Abbeel, and S. Levine, “Model-agnostic meta-learning for fast adaptation of deep
networks,” 2017.

指導教授

施國琛

審核日期

2025-1-9

推文