Study On Individualized Head-Related Transfer Function

In recent years, spatial sound becomes a research hot, with the rapid development of virtual reality technology. Binaural spatial sound can present spatial positions of sound source accurately, using only two recording channels, which are filtered by Head-Related Transfer Function (HRTF) of left/right ear. HRTFs are a set of filters from sound source to eardrums of human, which contain sound's spatial position cues. HRTF is measured through experiment, every person has different HRTF, because of different head, torso, shoulder, pinna etc. Measuring different people's HRTFs are expensive, generally experimenters measure a KEMAR's HRTF, which cause front-back confusion, up-down confusion, angular deviation etc and affect the reconstruction quality of spatial sound field.To remesy the above mentioned problems, the HRTF individualized method is studied. The main contributions in the thesis are as follows:(1)Use a structural model to model distance, azimuth, elevation separately, forming a cascade connection, based on the Peking University's (PKU) HRTF database. The elevation model parameters are calculated from PKU's Pinna-Related Transfer Function (PRTF). Through auditory feedback, a listener uses an interactive interface to tune azimuth parameters to change Interaural Time Delay (ITD) and head shadow effects; to tune spectrum peaks'and valleys'center frequency,3dB band with and center frequency gain to improve elevation locating. Finally, generate individualized HRTF with the tuned parameters.(2)Use Principal Component Analysis (PCA) with CIPIC HRTF database, and make PCA on horizontal plane'Head-Relate Impulse Response (HRIR) and median plane's Pinna-Relate Impulse Response (PRIR). At each angular, through auditory feedback, a listener uses an interactive interface to tune the first five Principal Component Weights (PCWs) with the largest standard deviations to improve location precision. Distance model, individualized horizontal HRIR and median PRIR are arranged in a cascade style to generate individualized HRTF. Futher more, for elevation locating, make PCA on PRTF's magnitudes.The PCWs and anthropometry parameters have Multiple Linear Regression (MLR) relation. The individualized PRTFs'magnitudes can be estimated by a person's anthropometry parameters and MLR weights. The phases are estimated from the Hilbert transform of PTRFs'magnitudes. The individualized PRIR is the Fourier inversion of PRTF.