Voice control for a gripper using Mel-Frequency Cepstral Coefficients and Gaussian Mixture Models

This work presents an implementation of a speaker-dependent speech recognition system used to control a gripper. The application was made using MATLAB and the gripper was assembled using the Lego Mindstorm NXT robotic kit. Four commands are implemented for controlling the gripper: Open, close, rotate left and rotate right. The development was divided into two stages. In training stage, we use Mel Frequency Cepstral Coefficients (MFCCs) and Gaussian Mixture Models (GMMs) to generate a representation of each defined command. Then, in testing stage, those models are used to identify the speaker´s utterance and send the command to the actuator. Finally, we present test results that show a performance of 95.09% for our system, and then we compare it with similar works.