A Biological Inspired Robotic Auditory System Based on Binaural Perception and Motor Theory

In this paper we present a novel artificial auditory system for humanoid robots. We address the problem of estimating an articulatory representation of the speech of the talker who is speaking to the robot using our auditory system. According to the motor theory of perception, the articulatory representation is the first step of a robust speech understanding process. The system is composed by two parts, namely a beamforming module and a perception module. The beam-former is two-channel (i.e. dual-microphones) and it is based on the super-directive beam-forming algorithm. The environment is scanned for seeking a sound source; when the direction of the source is found, the reception lobe of the dual-microphone system is steered to that direction and the signal is acquired. The perception module is based on a fuzzy computational model of human vocalization. In summary, the relationships between places of articulation and speech acoustic parameters are represented with fuzzy rules. Starting from the articulatory features, a set of acoustic parameters are generated according to the fuzzy rules. These acoustic parameters are used to generate a synthetic utterance which is compared in the perceptual domain to the corresponding spoken utterance. The goal of that is to estimate the membership degrees of the articulatory features using analysis-by-synthesis and genetic optimization. Introduction It is well known that through the auditory system, living ceatures gather important information about the world in which they live. For lower animals, it may mean to be able to escape from a danger or to catch a prey, for humans it may mean to be able to focus one’s attention on events, such as phone ringing, person talking etc. Robots also greatly benefit from auditory capabilities because their intelligence can be improved by fusing auditory information with the information coming from other sensors such as vision. The aim of this paper is to propose an artificial auditory system that gives a robot the ability to locate sounds sources using binaural perception, and to perceive speech, in terms of articulatory representation, on the basis of the motor theory of perception (Liberman & Mattingly 1985). In Fig. 1 the block diagram of our auditory system is reported. In summary, this paper focuses on the following two auditory capabilities Copyright c © 2006, American Association for Artificial Intelligence (www.aaai.org). All rights reserved. • binaural localization of human talkers • speech perception by articulatory features estimation