Evolving Emotional Behaviour for Expressive Performance of Music

Abstract

Today computers can be programmed to compose music automatically, using techniques ranging from rule-based to evolutionary computation (e.g., genetic algorithms and cellular automata). However, we lack good techniques for programming the computer to play or interpret music with expression. Expression in music is largely associated with emotions. Therefore we are looking into the possibility of programming computer music systems with emotions. We are addressing this problem from an A-Life perspective combined with recent discoveries in the neurosciences with respect to emotion.

Antonio Damasio refers to the importance of emotions to assist an individual to maintain survival, as they seem to be an important mechanism for adaptation and decision-making. Specifically, environmental events of value should be susceptible to preferential perceptual processing, regarding their pleasant or unpleasant. This approach assumes the existence of neural pathways that facilitate survival. Stable emotional systems should then emerge from self-regulatory homeostatic processes.

We implemented a system consisting of an agent that inhabits an environment containing with a number of different objects. These objects cause different physiological reactions to the agent. The internal body state of the agent is defined by a set of internal drives and a set of physiological variables that vary as the agent interacts with the objects it encounters in the environment. The agent is controlled by a feed-forward neural network that integrates visual input with information about its internal states. The network learns through a reinforcement-learning algorithm, derivate from different body states, due to pleasant or unpleasant stimuli.

The playback of musical recordings in MIDI format is steered by the physiological variables of the agent in different phases of the adaptation process.

The behaviour of the system is coherent with Damasio’s theory of background emotional system. It demonstrates that specific phenomena, such as body/world categorization and existence of a body map, can evolve from a simple rule: self-survival in the environment. Currently, we are in the process of defining a system of higher-level emotional states (or foreground system) that will operate in social contexts; i.e., with several agents in the environment reacting to objects and interacting with each other.