Accelerated Method for the Reduced-Parameter Modeling of Head-Related Transfer Functions for Customizable Spatial Audio KENNETH JOHN FALLER II 1 , ARMANDO BARRETO 1 , NAVARUN GUPTA 2 and NAPHTALI RISHE 3 Electrical and Computer Engineering Department 1 and School of Computing and Information Science 3 Florida International University Miami, FL 33174 USA Department of Electrical and Computer Enginering 2 University of Bridgeport Bridgeport, CT 06604 USA Abstract: - Many spatial audio (“3D sound”) systems are based on the use of Head-Related Transfer Functions (HRTFs). Since the measurement of these HRTFs for each prospective listener is impractical in many applications, developers frequently use “generic” HRTFs (e.g., from a mannequin), sacrificing the superior spatialization that could be provided by “individual” HRTFs. This paper presents an improved method for decomposing the impulse response of measured HRTFs, known as Head-Related Impulse Responses (HRIRs), into multiple delayed and scaled damped sinusoids, as a means to obtain the parameters that will instantiate a general model in order to produce the same impulse response as the original HRTF. This is the first step in developing alternative functional models of HRTFs that would contain only a few parameters related to anatomical features of the intended listener, which could be estimated for each user, i.e., “customizable” HRTF models. The new decomposition algorithm is based on the use of higher-order (higher than second order) Steiglitz-McBride functional approximations supplemented by frequency-domain selection of the most appropriate damped sinusoid to represent each of the components of the original, measured HRIR under analysis. Key-Words: - Head-Related Impulse Responses (HRIR), Prony modeling method, Steiglitz-McBride iterative approximation method, customizable spatial audio. 1 Introduction The proliferation of computer applications in many aspects of our daily lives has brought about an expanded field of applications for digital audio. In addition to its high fidelity, digital audio may be added with a sense of sound source location, managed independently as one more attribute of the digital sound. This constitutes the basis of three dimensional (3-D) spatial audio and its applications have become increasingly popular in scientific, and commercial systems. The spatial audio effect can be achieved through different methods. The multi-channel approach requires that speakers be physically positioned around the listener (e.g., Dolby® 5.1 array). However, this approach is not portable and, therefore, it is impractical for several applications. The two-channel approach applies digital signal processing (DSP) techniques to an original digital sound to create a pair of signals (left channel and right channel) that can be delivered to a listener through headphones. In consequence, this approach is much more portable. The DSP-based creation of the left and right channel signals (i.e., a “binaural sound”), involves the use of special filters, characterized by their impulse responses, which are known as Head-Related Impulse Responses (HRIRs). The transfer functions of these filters are known as Head- Related Transfer Functions (HRTFs), and are meant to emulate the effect of anatomical (torso, head, external ear, etc.) and environmental (walls, floor, etc.) factors which cause modification of a sound as it propagates from its source to each of the listener's eardrums. Therefore, every position and each ear will have a specific HRIR. Convolving a sound signal with the two HRIRs corresponding to a specific source position results in a binaural sound (left channel, right channel) that, when played to a listener through stereo headphones will cause a perception similar to that of a sound emanating from the source location in question, specified by azimuth, elevation and distance (Fig. 1). Proceedings of the 5th WSEAS Int. Conf. on CIRCUITS, SYSTEMS, ELECTRONICS, CONTROL & SIGNAL PROCESSING, Dallas, USA, November 1-3, 2006 263