Development of Indonesian Speech Recognition with Deep Neural Network for Robotic Command

Abstract

Research on speech recognition for several languages has been shown significant improvement for seamless interaction between human and robot. In this study, a system to command assistant robot with Indonesian speech recognition using deep neural network (DNN) has been proposed. The DNN architecture created by convolutional neural networks (CNNs), max pooling, and fully connected layers. The experiments performed on a self-constructed dataset with training, validation, and testing data in 0.8:0.1:0.1 ratio. This network built using Keras (TensorFlow Backend) and the result shows 99.43% accuracy on testing data and 89.57% on actual condition.