Issue |
ITM Web Conf.
Volume 70, 2025
2024 2nd International Conference on Data Science, Advanced Algorithm and Intelligent Computing (DAI 2024)
|
|
---|---|---|
Article Number | 04027 | |
Number of page(s) | 7 | |
Section | AI and Advanced Applications | |
DOI | https://doi.org/10.1051/itmconf/20257004027 | |
Published online | 23 January 2025 |
Deep Learning in Music Generation: A Comprehensive Investigation of Models, Challenges and Future Directions
Computer Science, University of California, Davis, 95616, Davis, CA, United State
Corresponding author: xchkong@ucdavis.edu
Deep learning has made a lot of progress in the field of music generation. It now has powerful tools for both preserving traditional music and creating new, innovative compositions. This review explores various and recent deep learning models, such as Long Short-Term Memory (LSTM) networks, Transformer-based models, Reinforcement Learning (RL), and Diffusion-based architectures, and how they are applied to music generation. LSTMs effectively capture temporal dependencies, which are vital for producing coherent melodies and chord progressions. Transformer models, like MUSICGEN and STEMGEN, handle large amounts of data and dependencies efficiently, but they need a lot of computational resources. Reinforcement Learning models, such as MusicRL, combine human feedback to fine-tune AI-generated compositions based on the individual's preferences. Diffusion-based models, like MusicLDM, enhance audio fidelity, though real-time application remains a challenge. The objective of emotion-conditioned models, such as ECMusicLM, is to combine music with emotional cues so that the output has a stronger emotional resonance. However, each model faces its own set of limitations, such as computational inefficiency, data dependency, and challenges in capturing complex emotional nuances. Future research should focus on improving the computational efficiency of these models, expanding training datasets, and integrating more interactive, real-time systems.
© The Authors, published by EDP Sciences, 2025
This is an Open Access article distributed under the terms of the Creative Commons Attribution License 4.0, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.
Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.
Initial download of the metrics may take a while.