Voice Recognition on Humanoid Robot Darwin OP Using Mel Frequency Cepstrum Coefficients (MFCC) Feature and Artificial Neural Networks (ANN) Method

Abstract

This research presents an algorithm enabling humanoid robots to learn to identify the human voice. Mel Frequency Cepstrum Coefficients (MFCC) Feature and Artificial Neural Networks (ANN) techniques are used together in this example. The Mel Frequency Cepstrum Coefficients Feature extracts features and converts the audio signal into many parameters. An ANN is a collection of tiny processing units that mimic human neural networks' behavior. The ANN is similar to how humans learn by utilizing examples or supervised learning. A Neural Network is set up for a specific task, such as pattern recognition or data classification, and then modified through training. In biological systems, learning entails altering existing synaptic connections between neurons. In the case of the Neural Network, this is accomplished by adjusting the weight values that exist in each link from input, neuron, and output. The robot is eventually given the directions to go, stop, turn left, and turn right using the previously stated commands in the testing environment. The collected results demonstrate that the proposed technique is effective.