ICSES 2008 INTERNATIONAL CONFERENCE ON SIGNALS AND ELECTRONIC SYSTEMS KRAKÓW, SEPTEMBER 14-17, 2008 _____________________________________________________________________________________________________________________________________________________________________________________________________ Copyright © 2008 by Department of Electronics, AGH University of Science and Technology Individual HRTF Measurements for Accurate Obstacle Sonification in an Electronic Travel Aid for The Blind Abstract - The article presents a study of virtual sound source localization errors with the use of personalized head related transfer functions (HRTFs) in light of design of an electronic travel aid for the blind. The proposed device for aiding visually disabled individuals in independent mobility requires presenta- tion of spatial sounds, which lead to the need to construct a system for efficient individual HRTF measurement. Measurements were performed for 15 sighted and blind indi- viduals. Verification trials limited to the frontal hemisphere have shown that the localization of virtual sound sources can be performed with accuracy reaching average errors of 6.36º in azimuth and 9.47º in elevation. Keywords: HRTF, Travel Aid, blindness, sonification I. INTRODUCTION There has been a longstanding research carried out on sen- sory substitution systems that compensate for the lost vision among the blind. One of the first reported electronic travel aids (ETA) for the visually impaired was built by a Polish scientist Kazimierz Noiszewski. In 1897 he made “an artifi- cial eye” - a device that used Selenium cells to convert light to sounds of strength proportional to the average scene brightness. Today, there unfortunately has been no single ETA system developed that would be ubiquitously used by the blind. Nevertheless, the most notable efforts include obstacle detectors such as the Ultracane [1] or Lasercane [2], as well as the more complex Sonic Pathfinder [3] and KASPA [4]. Each of these systems uses some form of en- ergy emitted into the environment. The reflected signals form patterns that are converted into special sound codes corresponding to scene obstacles. Other class of ETA systems are environment imagers that offer, as opposed to point-like obstacle detectors, a more complete representation of the scene. These devices capture images of the scene and convert this information-rich data into some kind of “auditory image”. The vOICe, converts a greyscale images into sound patterns [5]. The Espacio Acus- tico Virtual (EAV) [6] use stereovision for scene depth re- construction. These ambitious attempts, however, with the latter implementing the HRTFs (Head Related Transfer Functions), encounter the barrier of perceptive capacity mismatch between the imaging sensors and the human audi- tory system. The listener is simply overloaded with continu- ous sound streams rendered from the video streams. In our research on the ETA system for the blind, the key concept is to segment out the major objects from the scene (obstacles) and use only them for sonification. The four main modules of the constructed system are: • the stereovision rig with a narrow base (approx. 8 cm) for 3D scene reconstruction, • algorithm for scene segmentation and parametric de- scription of obstacles, • procedure for synthesis of sound streams associated with selected scene objects, • HRTF filtering for spatialization of sounds reproduced in the headphones. Currently, we are experimenting with different sonification schemes by employing the main principles of the Breg- mann’s theory of sound streams. In the presented study we concentrate only on one aspect of the research that deals with the auditory module of the ETA, namely the use of HRTFs for sound spatialization. An inter- ested reader can find more details on the image processing modules in references [7,8] or visit our project web page at www.naviton.pl [9]. II. INDIVIDUAL HRTF MEASUREMENTS Head-Related Transfer Function (HRTF) is defined as the transmittance of the acoustic filter created by the human body, especially the torso, head and ear pinnae, measured for specific directions in space. It describes all the phenomena responsible for spatial hearing [10]. Implementing HRTFs in the audio processing path, allows to simulate those phenomena and create the illusion of spatial audio on stereo headphones. A. Trials with generic HRTFs During the early development of the ETA, generic HRTFs from the CIPIC Database [11] were used for each user; how- ever, first attempts to synthesize a spatial auditory scene were not satisfactory. During a trial with 15 volunteers (both Michal Pec Michal Bujacz Pawel Strumillo Andrzej Materka Institute of Electronics, Technical University of Lódź 211/215 Wólczańska, 90-924 Lódź, Poland michal_pec@o2.pl; bujaczm@p.lodz.pl; pstrumil@p.lodz.pl; materka@p.lodz.pl 235