Technology Research Of Three Dimensional Sound Source Based On Head Related Transfer Function

With the rapid development of three-dimensional movies, the three-dimensional audio and video technology has a broad market prospect. However, compared to three-dimensional movies,the development of the three-dimensional audio technology is relatively backward. The main factors that restrict the promotion are as follows. Firstly, could it efficiently product the personalized three-dimensional audio source. Secondly, could it effectively eliminate the crosstalk in reconstruction of the three-dimensional audio source using the loudspeaker. From the two aspects of three-dimensional audio production and reproduction, the paper does a series of studies.Firstly the paper starts to do research on three-dimensional audio production. The composition of CIPIC database is analyzed and the efficient production method is discussed. Firstly the composition of CIPIC database is introduced. To verify that HRTF contains location information, the paper finds the mapping relationship between the feature and location, used to establish the Gaussian process regression model, aimed to determine the location of the sound source. And the sound source position mean prediction errors are analyzed under the different characteristics and the different number of training samples. Then, using the characteristic parameters of CIPIC, based on the k nearest neighbor method, the HRTF is divided into two types including the big pinna and the small pinna. Then, for HRTF data from different directions of the same individual, HRTF resonance and reflection model are proposed, used to implement the efficient production of three-dimensional sound. By using non-negative Toeplitz matrix decomposition algorithm, the HRTF is decomposed into the convolution between the resonance response and orientation dependence of the reflection response, where the former is independent on direction and the latter is dependent on direction. For the reflection response, it is dealt with sparse algorithm to reduce the number of non-zero elements in the reflection response, to improve the efficiency of three-dimensional audio production. Under the two conditions such as L₁-NNLS and L₁-LS, the effects on HRTF data are compared.Secondly the paper does research on three-dimensional audio reproduction, the principle of crosstalk cancellation filter is analyzed, and the traditional crosstalk cancellation filter is improved. Due to the delay of discrete time Fourier calculation, non-causal factors are introduced into the whole system. This paper solves the problem in the frequency domain, using causal constraints, to eliminate non-causal factors. Due to the introduction of regularization, lag ringing phenomenon exists in the final binaural signal. The method of minimum phase regular, will change delayed ringing into advanced ringing, putting ringing effects into sound transmission process, to reduce the impact of the real voice signal delay.Finally, the three-dimensional sound source production software, including original source module, static three-dimensional sound source module, dynamic three-dimensional sound source module and parameters selection module, is designed for users to use conveniently and for user to complete perception experiments. In the real transmission environment and the simulated transmission environment, on 10 test persons, 10 kinds of parameters, a perception experiment, verifies the effects of different kinds of parameters on three-dimensional acoustic source location information.