Study On Minimum Phase Head Related Transfer Function Compression Method

Human have three dimensional auditory the same as vision. Besides sound intensity, pitch and timbre, human can also perceive direction and distance of sound source. This can impress listener the sense of space. Binaural acoustic technology can be utilized to reproduce the stereo. Binaural acoustic technology is implemented by convolving mono audio with HRTF (Head Related Transfer function), then play back with headphone or one pair of loudspeakers. HRTF is represented in time domain by HRIR (head related impulse response). This technology has been applied to auditory navigation for blind, flight training, and virtual reality.HRTF captures the filtering effect of human torso, head and pinna to a sound propagating from a specific spatial position to the eardrum. Using the measured HRTFs, human can create a vivid3D sound illusion. The measured HRTF dataset containing large number of HRTFs for all directions is huge which leads to a heavy burden on the memory requirement of virtual3D sound system. The storage will grow exponentially when storing more HRTF dataset. If HRTF database storage is reduced without changing its acoustic positioning performance, it will solve the problem of heavy burden on the memory requirement. This thesis studied on the amount of HRTF dataset storage reduction with linear and nonlinear dimensionality reduction method. The experimental results show, the amount of storage can be reduce to0.3%when keeping acoustic positioning performance of HRTFs. The content of this thesis mainly includes as follows,(1) Nonlinear dimensionality reduction of HRTF dataset has been done by using LLE (Locally Linear Embedding). Feature selection and interpolation algorithm in origin HRTF analyze method using LLE are modified. Besides, this analyze process is extended to two-dimensional space, then2-dimensional distribution of feature HRTF is obtained.(2) Use CP decomposition to reduce dimensionality of minimum pahse HRTF dataset.(3) Use the existing2D-CFD (2-dimension Common Factor Decomposition) to reduce dimensionality of minimum pahse HRTF dataset; Then common acoustic pole IIR modeling based on least-squares method, iterative prefiltering, BMT(Balanced Model Truncation) are applied to the result of2D-CFD.(4) Combining the nonlinear dimensionality reduction and linear dimensionality reduction.