Speech Training System for Thai Therapeutic Robot

This paper focuses on development of speech training system for Thai children with speech disorder. The system is used with our therapeutic robot, named Chang Phood. It is a small humanoid robot that can recognize whether a child correctly pronounces one to three consecutive words predefined by a teacher from developed GUI. Chang Phood is an interactive robot that can provide responses to the child in order to motivate speech training. Speech feature is initially extracted in terms of Mel Frequency Cepstral Coefficient (MFCC) and trained by multilayer perceptron with backpropagation neural network. Chang Phood can recognize words with the average accuracy of 79.5%. It is tested with twelve typical children, seven boys and five girls, for investigating problems before clinical trials.