Presenter

Mr Georgios Athanasopoulos - ETRO-VUB [Email]

Abstract

Humans are known to perceive humanoid robots not as mechatronic devices but to attribute to them characteristics similar to those attributed to living organisms. Human-robot interaction is therefore expected to employ mechanisms similar to those of humans interacting with each other. One of these mechanisms is the ability of processing auditory information.
In autonomous human-robot interaction, auditory information can contribute to resolve complex problems such as the focus of attention, activity recognition, and more. Humanoid robots ultimately require an artificial audition module that allows the robot to process and interpret a combination of verbal and non-verbal auditory inputs. A key component of such a module is the acoustic localization. The acoustic localization not only enables the robot to simultaneously localize multiple persons and auditory events of interest in the environment, but also provides input to other auditory tasks such as speaker identification and speech recognition.
Over the years, various acoustic localization methods have been proposed. The use of microphone arrays in robots is an efficient and common approach to the localization problem. In real-life scenarios, humanoid robots are expected to operate in various types of environments exhibiting different, often dynamically variable, acoustic properties. Moreover, the artificial audition must be capable of coping with a number of limitations imposed by real-world robots, such as their shape or the number and quality of available microphones.
This dissertation, moving away from simulated environments, focuses on the problem of acoustic localization for humanoid robots under real-world conditions and limitations. We study the robot's shape and surface material influence on the location estimation and introduce a framework for addressing their effect. The proposed method is based on a two-dimensional set of pre-measured time delays, and experiments with the humanoid robot NAO confirm its capability over previous approaches. Moreover, a series of enhancements is introduced spanning from compensating the imperfect frequency response of the array microphones, to smoothing the acoustic localization output for eliminating estimation outliers and enhancing its performance.
Motivated by the importance of the signal's phase information in the acoustic localization, this dissertation also proposes different pre-processing techniques for improving the localization accuracy by enhancing the noise-distorted phase spectra of the array signals. Our evaluation shows the contribution of the proposed enhancements in the acoustic localization performance under joint noisy and reverberant conditions. Finally, experimental results with NAO demonstrate the capability of the acoustic localization system to locate multiple speakers in a real-world environment.

Short CV

Electrical and Computer Engineering, 2004, University of Patras
Postgraduate in Signal Processing for Communications and Multimedia, 2006, National and Kapodistrian University of Athens