Sound Localization of Elevation using Pinnae for Auditory Robots

Abstract

Humans perceive sound lags and differences in loudness between their two ears to ascertain the location of sounds in terms of azimuth, elevation and interval, etc. Humans and animals can determine the direction of sounds using only two ears, by making use of interaural time difference (ITD), interaural phase difference (IPD), interaural intensity difference (IID), etc. Robots can also use sound localization to detect changes in the environment around them; consequently, various sound localization methods for robots have been investigated. Since many animals have two ears, two ears seem to be the minimum requirement for determining sound localization. While most robot’s sound localization systems are based on microphone arrays that consist of three or more microphones, several researchers have attempted achieving binaural sound localization in robots. Nakashima (Nakashima and Mukai, 2004) developed a system that consisted of two microphones with pinnae and a camera that was only utilized in learning. By adopting a neural network that estimated the control command, he proposed a method utilizing sound signals for guiding the robot in the direction of a sound. Takanishi (Takanishi et al., 1993) achieved continuous direction localization of a sound source in the horizontal plane. They used the interaural sound pressure difference and the ONSET time difference as parameters and proposed a method for achieving two-step direction localization in a humanoid head using these parameters. Kumon (Kumon et al., 2003) approached the problem using an adaptive audio servo system based on IID as the control method for orientating the robots in the direction of the sound source in the horizontal plane by using two microphones. This chapter describes the use of an audio servo of elevation for achieving sound localization using spectral cues caused by pinnae. Here, the term “audio servo” denotes a method for simultaneously localizing sound sources and controlling the robot’s configuration, in contrast with conventional methods that are based on the “listen-and-move” principal. The audio servo is characterized by a feedback system that steers the robot toward the direction of the sound by combining dynamic motion of the robot with auditory ability. In order to achieve this, this chapter proposes a method for detecting instantaneous spectral cues, when these cues are not very accurate. Furthermore, controllers that compensate for the inaccuracy of the measured signalare also considered. In addition, this