| Abstract [eng] |
Current artificial intelligence (AI) tools for music creation are pre-trained using general datasets. As a result, AI models assimilate general principles of music creation, while the unique style of individual artists is not highlighted. This can lead to stylistic homogenization of compositions and raise copyright concerns. This issue is particularly relevant for creators seeking to integrate AI tools into their creative process while maintaining a distinctive sound. This study aims to investigate how to personalize the training of AI models with limited resources by using personal creative content, as well as how to creatively adapt existing tools trained on general data. This dual perspective seeks to provide creators with methods to effectively integrate AI tools into the creative process without losing originality and artistic authenticity. Research objective – the use of AI tools in the music creation process and the possibilities of training generative AI (GAI) with limited resources. Project goal – to explore the potential of AI in original music creation and to develop a simplified GAI training methodology for music creators without programming experience. Tasks: 1. Examine the possibilities offered by pre-trained AI for original music creation. 2. Analyze available GAI training options, considering existing resources (financial, technological, and knowledge-based). 3. Define the concept of original music in the context of AI. 4. Apply the analyzed pre-trained AI capabilities and identified tools to original music creation to test AI’s creative potential in practice. 5. Implement the analyzed GAI training options and tools for original music creation to test the creative possibilities of trained GAI in practice. An experiment was conducted using "Magenta" models ("MelodyRNN" and "MusicVAE") trained on personal compositions, as well as a Markov chain model through "Max for Live." Additionally, pre-trained AI tools such as "Suno" and "Udio" were employed, with their generated ideas creatively integrated into compositions. The results demonstrated that personalized AI models can generate new musical ideas while preserving the artist's style, however, they are less flexible than the most advanced AI music generators and pose technical challenges. Pre-trained models provide interesting interpretations, but their use limits control and raises copyright issues. This work highlights the potential and limitations of AI in the creative process, opening opportunities for new experiments, rhythms, and melodies while preserving artistic authenticity. |
