Vocal Signal Enhancement for Measurement of Child Robot Interaction

This paper presented data prepossessing approaches to reduce the ambient noise and interference for assessing young children's engagement in collaborative activities mediated by a humanoid robot. During the data collecting session, children's speech signals have been recorded. In order to extract meaningful features for identify children's engagement levels, the ambient noise has been removed by using speech enhancement technique. In this research, we applied band pass filter, integrative wiener filter and adaptive noise cancellation to improve the speech quality. We proposed to apply an innovative double adaptive noise cancellation method to cancel the unwanted noise. Experimental results show that the proposed method can reduce the noise dramatically. Then, the engagement level can be recognized based on clean speech signals.