Vowel Acquisition based on Visual and Auditory Mutual Imitation in Mother-Infant Interaction

A pioneering constructivist approach to building a robot that reproduces a developmental process of infants’ vowel acquisition has been conducted by Yoshikawa et al. [1] inspired by the observation in infant study. They have constructed a mother-infant interaction model with robot learning capability and parrot-like teaching by caregiver. However, the robot has not listened his/her own voice, therefore it could not actively explore more natural vowels similar to the caregiver. The study presented in this paper extends the previous work in the following manners seeking for more natural interaction. First, a lip is added to the robot to imitate the lip shape of the caregiver in order to accelerate the learning process by constraining the initial exploration area in the formant space. Second, the pentagon the caregiver’s vowels construct in the formant space is utilized as the desired vowels for the robot. Third, mutual imitation between the robot and the caregiver is introduced in order to obtain more natural vowels hypothesizing that the caregiver imitates the robot voice but unconsciously the imitated voice is close to one of his/her own vowels. Through this process, the desired positions specified at the second step is gradually shifted, expecting to more natural ones. The experimental results are shown and the future issues are discussed.