Customization Of Head-related Transfer Functions Based On Anthropometric Parameters

Head-related transfer functions (HRTFs) is a key part of virtual auditory and individualHRTFs ensure authentic and nature auditory effect in field of virtual auditory. Due tolimitations of traditional measurement and computation, customization of individual HRTFsbecomes research hotspot. This paper presents a series of individualized method of HRTFsbased on anthropometric parameters, and objective evaluation and subjective experiment areconducted to testify their effectiveness.We first propose a measurement of anthropometric parameters based on3D scannedhead model and KEMAR is used as an example to illustrate the effectiveness of this newmethod. The result indicates that the relative deviation of the data of new method and theoriginal design data is less than1%for most anthropometric parameters. This work providesreliable data foundation for subsequent research.Additionally, from view of selecting anthropometric parameters, we advanceanthropometric parameters matching method of individual HRTFs. We choose fouranthropometric parameters group from individual data of35subjects in CIPIC HRTFdatabase by specific selecting process. The computing result shows that new matchingmethod is better than Zotkin’s method with2dB decrement on spectral distortion (SD) formost evaluating subjects, besides, the subjective auditory experiment demonstrates theadvancement of our proposed matching method that has new anthropometric parametersgroup on auditory aspect.In a further study, we propose anthropometric parameters prediction of individualHRTFs. Firstly, we use Independent Component Analysis (ICA) to decompose HRTFs,HRTFs of KEMAR in25directions of median plane are selected as an example to decomposeand reconstruct by ICA. Both SD and subjective experiment illustrate that SDs of six orderdecomposition of HRTFs are less than2dB from0degree to45degree in median plane,distortion of reconstruction does not bring any auditory perception. But other high and lowazimuths in median plane, the order of HRTF reconstruction need to be above than six inorder to satisfy the auditory requirement. According to mentioned conclusions, on the onehand, HRTF ICA is used to decompose HRTFs of35CIPIC HRTF subjects in median plane,obtaining individual weight coefficient. On the other hand, factor analysis is used to decideanthropometric parameters that regression model requires. Last but not least, we use multiplelinear regression to analyze mathematic relation between weight coefficient andanthropometric parameters, and establishing anthropometric parameters prediction model of individual HRTFs based on ICA. For most azimuths in median plane, SDs and experimentimply that predicting HRTFs are better than non-individual (KEMAR) HRTFs.This essay studies customization of individual HRTF based on anthropometricparameters systematically. This research achievement is not only beneficial to understandingcomplicated relationship between HRTFs and anthropometric parameters, but also widelyapplies on real virtual auditory system.