Title:Self-supervised reinforcement learning for speaker localisation with the iCub humanoid robot

Abstract:In the future robots will interact more and more with humans and will have to communicate naturally and efficiently. Automatic speech recognition systems (ASR) will play an important role in creating natural interactions and making robots better companions. Humans excel in speech recognition in noisy environments and are able to filter out noise. Looking at a person's face is one of the mechanisms that humans rely on when it comes to filtering speech in such noisy environments. Having a robot that can look toward a speaker could benefit ASR performance in challenging environments. To this aims, we propose a self-supervised reinforcement learning-based framework inspired by the early development of humans to allow the robot to autonomously create a dataset that is later used to learn to localize speakers with a deep learning network.