Deep learning Models Used to Generate the Music in Natural Language Process

For the goal of generating art, including music, deep learning-based language models shown encouraging outcomes. However, because to the subjective nature of music, it is difficult to put a numerical value on how well a performance sounds. As a result, most evaluations of symbolic musical generation models rely on simplistic mathematical measures, such the loss function's output. Using just creation of classical piano music, the goal of this study was to quantify and assess musical samples generated by models of deep learning from a human viewpoint. Using musical samples composed by humans and musical samples produced by artificial intelligence models, such as Performance RNN, Music Out of the 118 people who took part in the assessment, 118 were blind tested using Transformer, Muse Net, & a GRU-based proprietary model. During the experiments, it was shown that musical samples created using models based on the architecture of the Transformers neural networks were the most responsive to the test population, surpassing human compositions. In addition, the results demonstrated that participants' ability to identify the compositional source of audible samples was significantly enhanced by their degree of musical sensitivity & competence.