| The main focus of this thesis is the application of contemporary signal processing and system modeling techniques to free-field-to-eardrum transfer function (FETF or HRTF) estimation, to modeling the acoustical behavior of the external ear, and to simulating virtual auditory environments.; External ear modeling work reported in the literature, primarily during the past 30 years, is summarized. We report the superiority of system identification approaches for FETF estimation over empirical DFT based methods. A beamforming model for the FETFs that characterizes the ear as a multi-sensor acoustic array is developed. Its modeling performance and limitations are discussed. A Spatial Feature Extraction and Regularization (SFER) model (also termed two-stage model) based upon Karhunen-Loeve expansion (KLE) of the FETFs and application of generalized smoothing splines is then proposed. The KLE is used to derive a low dimensional subspace in which the FETFs lie. The bases of this subspace are complex-valued eigen-transfer functions (EFs). By projecting all measured FETFs onto this EF set, the coordinates of each FETF in the low dimensional subspace are determined. These coordinates represent samples of extracted spatial features of the FETF. Functional representations of these spatial features termed Spatial Transformation Characteristic Functions (STCFs) are obtained by applying a thinplate generalized spline smoothing model to "regularize" the sample sets. Several relationships between STCFs and external ear geometry are noted. A functional representation for the FETF is obtained by linearly combining the EFs with the STCFs. The effectiveness of the SFER model is fully validated acoustically using a large body of experimental data. Typical errors between the measured and modeled FETFs for a KEMAR are on the order of hundredths of one percent.; The SFER model establishes a framework for both theoretical spatial hearing research and technical advancement of virtual environments. This model enables the synthesis of both spatially and temporally dynamic stimuli. We hope further research based on this model will lead to new developments in auditory neurophysiology, psychoacoustics, and the engineering of virtual auditory environments. |
