Imitation Learning Based Talking Heads in Humanoid Robotics

Abstract

The main goal of this Chapter is to describe a novel approach for the control of Talking Heads in Humanoid Robotics. In a preliminary section we will discuss the state of the art of the research in this area. In the following sections we will describe our research results while in the final part some experimental results of our approach are reported. With the goal of controlling talking heads in mind, we have developed an algorithm which extracts articulatory features from human voice. In fact, there is a strong structural linkage between articulators and facial movements during human vocalization; for a robotic talking head to have human-like behavior, this linkage should be emulated. Exploiting the structural linkage, we used the estimated articulatory features to control the facial movements of a talking head. Moreover, the articulatory estimate is used to generate artificial speech which is by construction synchronized with the facial movements. Hence, the algorithm we describe aims at estimating the articulatory features from a spoken sentence using a novel computational model of human vocalization. Our articulatory features estimator uses a set of fuzzy rules and genetic optimization. That is, the places of articulation are considered as fuzzy sets whose degrees of membership are the values of the articulatory features. The fuzzy rules represent the relationships between places of articulation and speech acoustic parameters, and the genetic algorithm estimates the degrees of membership of the places of articulation according to an optimization criteria. Through the analysis of large amounts of natural speech, the algorithm has been used to learn the average places of articulation of all phonemes of a given speaker. This Chapter is based upon the work described in [1]. Instead of using known HMM based algorithms for extracting articulatory features, we developed a novel algorithm as an attempt to implement a model of human language acquisition in a robotic brain. Human infants, in fact, acquire language by imitation from their care-givers. Our algorithm is based on imitation learning as well. Nowadays, there is an increasing interest in service robotics. A service robot is a complex system which performs useful services with a certain degree of autonomy. Its intelligence emerges from the interaction between data gathered from the sensors and the management algorithms. The sensorial subsystem furnishes environment information useful for motion tasks (dead reckoning), auto-localization and obstacle avoidance in order to introduce