Workflow Analysis and Improvements in Music Driven Dance Moves Generation

The broad and dynamic fluidity of human motion has always been difficult to understand from the view of computations. Decoding and improving on such a field is of great importance for research and the automation industry. Dance is the culmination of human art, culture and evolution. The expressiveness of the motion is a field of interest and generating such realistic motion through computer intelligent models is a great way of understanding the intricacies of human motion generation. Through this research project, we would like to propose improvements in the workflow of motion generation with just an audio input. The subjectivity of the art form enables us to further explore the workflow both quantitatively and qualitatively. Dance diversity has been a constant point of interest due to the possibilities of interpretation and subjectivity among humans. We focus on exploring a new approach of preprocessing audio files of varying tempo using “jukemirlib” and then continue with the “actor critic reward function” to include as much diversity in the generated output through our work all the while remaining in the context of music to dance with much emphasis given to visual evaluation.