Development of Speech Based Control System for Human Robot Interaction (HRI)

Abstract

Human Robot interaction (HRI) has recently received considerable attention because communication is a very important parameter to increase human efficiency in HRI. Integration of speech based control system in wheel chair robot introduces flexible interaction between the handicapped person and the wheel chair. Wheel chair robot increases the mobility of physically handicapped person, thus making them independent in doing their daily works. Here an algorithm is proposed for isolated word recognition based speech recognition system. In this algorithm after low pass filtering process the voiced part and the unvoiced part of the speech sample are separated using zero crossing rate and short term energy method. Then the derived features (modified MFCC features and LPC features) are extracted from the speech sample and are then finally stored in the database. Speech recognition is carried out using KNN and PNN based supervised learning methods. A real time prototype is developed using CATIA V5 CAD software and Stratasys dimension 3D printer using ABS plastic as prototyping material and Stratasys Objet 30 using veroclear rgd810 as prototyping material. Two DC motors, L293D motor driver and an Arduino Uno board are installed on prototype. This system is then tested using MatLab Arduino Interface.
Here four different combinations of feature extraction methods with classifier were tested resulting into 72.5%, 78.75%, 76.25% and 83.75% accuracy for combinations of LPCKNN, MFCC-KNN, LPC-PNN, and MFCC-PNN respectively.