LARRISON, ZIKORA (2026) DESIGN AND IMPLEMENTATION OF AN AI-BASED MUSIC GENERATION SYSTEM USING DEEP LEARNING. Other thesis, Godfrey Okoye University, Enugu.
|
Text
ZIKORA LARRISON FINAL PROJECT correction.pdf Download (1MB) | Preview |
Abstract
The AI music generation systems that are existing currently are struggling with a lot of limitations that are critical. This limitations include but are not limited to, a dataset bias that is is persuasive towards the traditions of western music, the insufficient control of users over creative parameters like the genre, tempo and emotions of the produced music, the high demand of computational power which in turn limits and restricts the accessibility of the system by the end users and also the metrics of evaluation that fail to capture the great aesthetics and dimensions of cultural music. this project work of research therefore presents the design and implementation of an AI based music generation system. This system that is presented by this project work therefore addresses these challenges directly by applying transformer architectures. The system is trained using datasets that are available publicly and therefore, makes use of the self-attention mechanism such as MAESTRO and Lakh MIDI. the system also integrates a user interface that is very user-friendly in other enable the granular customization of the music that is produced. At the end of the research, the system was evaluated experimentally and the result yielded a BLEU score of 0.72 and a perplexity of 18.97. the system was also tested comprehensively and the result showed that the system was able to generate compositions that are stylistically diverse and coherent. The system was able to do this while it maintained computational efficiency that is suitable for a hardware environment that is modest. This project work however was able to contribute to computational creativity by advancing the applications of transformers in the generation of music that is symbolic. The study also offered an evaluation framework that is more balanced and with that fostered greater collation between AI and humans in endeavors that are artistic.
| Item Type: | Thesis (Other) |
|---|---|
| Subjects: | Q Science > Q Science (General) |
| Divisions: | Faculty of Engineering, Science and Mathematics > School of Electronics and Computer Science |
| Depositing User: | MICHAEL MADUBUKO |
| Date Deposited: | 22 Jul 2026 14:40 |
| Last Modified: | 22 Jul 2026 14:40 |
| URI: | http://eprints.gouni.edu.ng/id/eprint/6049 |
Actions (login required)
 |
View Item |
