Gurjit Sangha Undergraduate Dissertation 2014/15

Active Hearing with the NAO Robot

Supervised by G.Brown

Abstract

This project aimed to develop a sound source localisation system for the NAO, a humanoid robot made by Aldebaran Robotics. Two subsystems were implemented, one to process the audio data received by the robot, the other to process the visual data. The audio process would use cues such as the Interaural Time Difference (ITD) of the sound signal as well as techniques including head rotation to turn to face where it had determined the sound source to be. The visual system would then detect any faces in the robot's view and turn so that they are in the centre of it.

The system was tested using a series of evaluation tests. The results showed that it took on average just over one attempt to find the correct sound source location when the distance between the source and the robot was half a metre and there was minimal background noise. It was found that the number of attempts increased as the distance and noise level increased. When multiple sound sources were present, it took the system just under three attempts to find them all.