Creating Music with Artificial Intelligence (Transformer Algorithm)

Creating Music with Artificial Intelligence (Transformer Algorithm)

Alireza Rajabi1 Rouhollah Arjomand2

1) M.Sc. in Computer Software, Department of Computer Engineering, Islamic Azad University, Abadeh Branch, Iran
2) Supervisor, M.Sc. in Computer Software, Department of Computer Engineering, Islamic Azad University, Abadeh Branch, Iran

IOI Code : ABEB-ACAFFG
Publication : 4th International Conference on Recent Advances in Engineering, Innovation and Technology - Belgium(eitconf.com)
Abstract :
    The expansion of artificial intelligence in recent years has transformed the methods used to generate, process, and analyze music. The central question of this research concerns how computational systems—relying on data and mathematical patterns—can generate novel musical structures, and what capabilities and limitations characterize this process. This paper aims to review and analyze technical methods for music generation, compare symbolic and audio-based approaches, and detail a proposed algorithm based on an autoregressive Transformer architecture. The research employs a review-analytical methodology, drawing upon scientific literature, technical documentation, and established datasets within the field of computational music. The study examines various techniques, including Recurrent Neural Networks (RNNs), Long Short-Term Memory (LSTM) networks, Generative Adversarial Networks (GANs), Variational Autoencoders (VAEs), Transformers, and diffusion models. An analysis of the literature reveals that each method strikes a different balance between sequence quality, controllability, computational cost, and long-term coherence. The proposed algorithm converts musical events into tokens and employs autoregressive prediction to generate the next token; this enables the creation of symbolic sequences, which can subsequently be converted into MIDI files. The overall conclusion is that artificial intelligence serves as an effective tool for composition and ideation; however, the artistic quality of the output remains contingent upon training data, model design, human evaluation, and the composer s oversight.

Keywords : artificial intelligence music generation deep learning transformer symbolic music