MUSIC GENERATION BASED ON MOOD THROUGH UTILIZATION OF BIAXIAL LSTM ARCHITECTURE

• Main Author: Anjas Prabowo, Paskahlis
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/39322
• Institution: Institut Teknologi Bandung

Artificial intelligence has been showing a rapid advancement since the term was first introduced in 1952 by John McCarthy et al. The application of artificial intelligence to human life also varies, from security, health, to entertainment/art, especially music. Nowadays there are many virtual composers, which is one form of application of artificial intelligence to generate music based on certain references. DeepJ is one of several virtual composers that currently exists. The research aims to utilize the Biaxial LSTM architecture, which was adapted from DeepJ, to build a model that is capable of producing music that reflects certain combinations of moods referring to Thayer's Mood Model. In this case, a set of MIDI format music is used with each mood label as a dataset. Based on experiments, evaluations, and analyzes that has been carried out, the model is proven to be able to generate music that reflects a combination of certain moods. Even so, the results of the generation of music that reflects anxious mood are not optimal yet. That lack of music quality is suspected to be caused by the lack of quality of the dataset and number of epoch training. Thus, for the purposes of further development or research, improvements can be made to the quality of datasets, the addition of the number of epoch training, as well as integration with other systems.