Navarun Gupta & Armando Barreto International Journal of Scientific and Statistical Computing (IJSSC), Volume (1): Issue (2) 20 A Customizable Model of Head-Related Transfer Functions Based on Pinna Measurements Navarun Gupta navarung@bridgeport.edu Assistant Professor, Department of Electrical Engineering University of Bridgeport Bridgeport, CT 06604, USA Armando Barreto barretoa@fiu.edu Associate Professor, Department of Electrical Engineering Florida International University Miami, FL 33174, USA Abstract This paper proposes a method to model Head-Related Transfer Functions (HRTFs) based on the shape and size of the outer ear. Using signal processing tools, such as Prony’s signal modeling method, a dynamic model of the pinna has been obtained, that completes the structural model of HRTFs used for digital audio spatialization. Listening tests conducted on 10 subjects showed that HRTFs created using this pinna model were 5% more effective than generic HRTFs in the frontal plane. This model has been able to reduce the computational and storage demands of audio spatialization, while preserving a sufficient number of perceptually relevant spectral cues. Keywords: HRTF, Binaural, HRIR, Pinna, Model 1. INTRODUCTION HRTFs represent the transformation undergone by the sound signals, as they travel from their source to both of the listener’s eardrums. This transformation is due to the interaction of sound waves with the torso, shoulder, head and outer ear of a listener [9]. Therefore, the two components of these HRTF pairs (left and right) are typically different from each other, and pairs corresponding to sound sources at different locations around the listener are different. Furthermore, since the physical elements that determine the transformation of the sounds reaching the listener’s eardrums (i.e., the listener’s head, torso and pinnae), are somewhat different for different listeners, and so should be their HRTF sets [2]. Currently, some spatialization systems make use of HRTFs that are empirically measured for each prospective user. These “custom” HRTFs are anthropometrically correct for each user, but the equipment, facilities and expertise required to obtain these “measured HRTF pairs”, constrain their application to high-end, purpose- specific sound spatialization systems only [2]. For most consumer-grade applications, sound spatialization systems resort to the use of “generic” transfer functions, measured from a manikin with “average” physical characteristics [7], which, evidently is a fundamentally imperfect approach. This paper reports on our work to advance an alternative approach to sound spatialization, based on the postulation of anthropometrically-related “structural models” [6] that will transform a single-channel audio signal into a left/right binaural spatialized pair, according to the sound source simulation. Specifically, the work reported here proposes linkages between the parameters of the HRTF model and key anthropometric features