The authors have developed a vocalization training system for the auditory impaired using a talking robot. The training system mainly consists of a talking robot that has mechanical organs like a human. With an adaptive learning strategy using an auditory feedback control, the robot autonomously learns the vocalization, and then reproduces the speech articulation from inputted sounds. In the previous system, the speech-learning algorithm of the robot was constructed by employing a self-organizing neural network (SONN), which consists of the combination of a self-organizing map (SOM) and a neural network (NN). However, improper maps were occasionally generated in the results of the speech articulation learning. To solve this problem, a new algorithm introducing two three-dimensional SOMs, called a dual-SOM, was employed for the autonomous learning of the robotic articulations. By applying the robot and its properties, we have constructed an interactive training system. The training is divided into two approaches; one is to use the talking robot to show the shape and the motion of the vocal organs, and the other is to use a topological map to present the difference of phonetic features of a trainee's voices. In this study, first, the construction of the training system is described together with the autonomous learning of robotic vocalization using the dual-SOM algorithm, and then the analysis of the speech training progress is presented based on the phonetic features and the mechanical vocal articulations.
|ジャーナル||International Journal on Disability and Human Development|
|出版ステータス||Published - 2011 3|
ASJC Scopus subject areas