2D Binaural sound localization

  • Ahmed Rashed Kulaib AlTunaiji

Student thesis: Master's Thesis

Abstract

Being able to localize the position of a sound source is an important issue in robotics and many other application areas since it enables those systems to interact with the environment. For example, USAR robotics can use sound to search for hidden victims that are shouting for help. This thesis covers the problem of 2D sound localization using only two microphones (binaural). The three models used in this thesis are inspired by the human auditory system and are based on Interaural Time Di®erence (ITD), Interaural Level Di®erence (ILD), and Head Related Transfer Function (HRTF). The ITD model relies on the di®erence of arrival times of the sound signal be- tween the two microphones. The ITD model is used to measure the direction of a sound source in the horizontal plane. In this thesis it has been tested under various conditions and its accuracy and repeatability have been measured. It was found that accuracy degrades from 97% to 68% if a barrier such as a head is placed between the microphones due to re°ection and di®raction e®ects. A method of calibration for these e®ects was investigated and implemented, increasing the accuracy up to 97%. ILD can also be used to localize the direction of a sound source in the horizontal plane. ILD is de¯ned as the di®erence in the energy level between the two micro- phones. In this thesis the ITD and ILD models are combined together in order to measure the distance to the sound source. It has been shown theoretically that this model can measure the distance if the microphones are at least 1m apart, which is not practical for USAR environments. This is because the measurement resolu- tion is not su±cient to discriminate reliably between the level di®erences when the interaural distance is small. The HRTF is a spectral response that is formed when the sound waves hits the head and it contains notches that change position depending on elevation angle. To easily and accurately extract notches we used a spiral shaped pinna that allows the position of notches in the frequency spectra to change linearly as a sound source moves in the vertical plane. This allows us to form a database which can be used as a lookup table to detect the direction of a sound source in the vertical plane. The error in the estimated elevation angle is mostly less than 5 degree. These models have been used together to measure the direction of a blind sound source (unknown spectral characteristics) in 2D planes using only two microphones.
Date of Award2009
Original languageAmerican English
SupervisorDavid Vernon (Supervisor)

Keywords

  • 2D Binaural Sound
  • Localization

Cite this

'