| Resumo: | The mid-20th century witnessed the emergence of an area of study that focused on the automatic generation of musical content by computational means. Early examples focus on offline processing of musical data and recently, the community has moved towards interactive online musical systems. Furthermore, a recent trend stresses the importance of assistive technology, which pro-motes a user-in-loop approach by offering multiple suggestions to a given creative problem. In this context, my research aims to foster new software tools for creative support systems, where algorithms can collaboratively participate in the composition flow. In greater detail, I seek a tool that learns from variable-length musical data to provide real-time feedback during the composition process. In light of the multidimensional and hierarchical structure of music, I aim to study the representations which abstract its temporal patterns, to foster the generation of multiple ranked solutions to a given musical context. Ultimately, the subjective nature of the choice is given to the user to which a limited number of 'optimal' solutions are provided. A symbolic music representation manifested as Multiple Viewpoint Models combined with the Variable Markov Oracle (VMO) automaton, are used to test optimal interaction between the multi-dimensionality of the representation with the optimality of the VMO model in providing both style-coherent, novel, and diverse solutions. To evaluate the system, an experiment was conducted to validate the tool in an expert-based scenario with composition students, using the creativity support index test.
|
|---|