Intelligent speech control system for human-robot interaction

Abstract

Accurately extracting subjective contents of speech signals and applying it on controlling robots remain to this day a challenging task as well as an insistent demand in human-robot interaction (HRI). A simple classification of human's intentions may limit the development of robots' natural reactions to users. Additionally, there should be a control system that can understand and translate human's intentions into control inputs. This paper proposes an intelligent speech control system for HRI. The objective is to understand human' s speech commands via recognizing, quantifying audio signals and translating speech inputs into control inputs. Aiming at this purpose, three main parts for the system are designed: a speech recognition system, a speech measurement system and a control system. Specifically, Mel Frequency Cepstral Coefficient (MFCC) and Dynamic Time Warping (DTW) techniques are utilized to recognize isolate speech commands. An energy-based feature and novelly proposed spectrum-based features are introduced to represent subjective contents of speech signals followed by Random Forest (RF) as a regressor. Several control schemes are utilized to translate quantified speech signals into control inputs. Simulation results illustrate the performance of the proposed system and the robot adaptive control system outperforms other control methods on effectiveness and controllability. The improved spectrum-based features demonstrate the capacity to extract subjective information of signals.