Frontiers in Neurorobotics (Apr 2013)
Reward-Based Learning for Virtual Neurorobotics Through Emotional Speech Processing
Abstract
Reward-based learning can easily be applied to reallife with a prevalence in teaching methods for children. It alsoallows machines and software agents to automatically determinethe ideal behavior from a simple reward feedback (e.g. encouragement) to maximize their performance. Advancements inaffective computing, especially emotional speech processing (ESP)have allowed for more natural interaction between humans androbots. Our research focuses on integrating a novel ESP systemin a relevant virtual neurorobotic application. We created anemotional speech classifier that successfully distinguished happyand sad utterances. The accuracy of the system was 95.3%and 98.7% during the offline mode (using an emotional speechdatabase) and the live mode (using live recordings), respectively.It was then integrated in a neurorobotic scenario, where a virtualneurorobot had to learn a simple exercise through reward-based learning. If the correct decision was made the robotreceived a spoken reward, which in turn stimulated synapses (inour simulated model) undergoing spike-timing dependent plas-ticity (STDP) and reinforced the corresponding neural pathways.Both our emotional speech processing and neurorobotic systemsallowed our neurorobot to successfully and consistently learnthe exercise. The integration of ESP in real-time computationalneuroscience architecture is a first step toward the combinationof human emotions and virtual neurorobotics.
Keywords
