Generative models for music using transformer architectures

This thesis focus on growth and impact of Transformes architectures which are mainly used for Natural Language Processing tasks for Audio generation. We think that music, with its notes, chords, and volumes, is a language. You could think of symbolic representation of music as human language. A brief sound synthesis history which gives basic foundation for modern AI-generated music models is mentioned . The most recent in AI-generated audio is carefully studied and instances of AI-generated music is told in many contexts. Deep learning models and their applications to real-world issues are one of the key subjects that are covered. The main areas of interest include transformer-based audio generation, including the training procedure, encoding and decoding techniques, and post-processing stages. Transformers have several key advantages, including long-term consistency and the ability to create minute-long audio compositions. Numerous studies on the various representations of music have been explained, including how neural network and deep learning techniques can be used to apply symbolic melodies, musical arrangements, style transfer, and sound production. This thesis largely focuses on transformation models, but it also recognises the importance of numerous AI-based generative models, including GAN. Overall, this thesis enhances generative models for music composition and provides a complete understanding of transformer design. It shows the possibilities of AI-generated sound synthesis by emphasising the most current developments.