Research On Spherical Microphone Array Recording And Binaural Virtual Rendering System

Binaural virtual rendering of signals captured by spherical microphone array is a newly-developed method for recording and reproducing spatial information of a sound field. Based on the outputs of spherical microphone array, binaural signals are synthesized dynamically and then reproduced via headphones or loudspeakers. This method is able to reconstruct the spatial information of the sound field exactly with great universality and flexibility, thus is applicable to hearing research, multimedia, virtual reality, etc.Although binaural virtual rendering of signals captured by spherical microphone array has been proposed previously, it is still under developed and there are many aspects to be improved. Up to now, there are only several research groups in the world and no other groups in China who are able to reconstruct the whole system including capturing and reproduction. The signal processing in the system also needs to be optimizted. Moreover, no reports are found in the references which deal with the virtual source localization experiment on the system. To address these problems, this dissertation has carried out the following studies:Firstly, the effect of the methods for measuring the impulse responses of spherical microphone array is systematically analyzed. Particularly, the difference among the results from various measuring methods was interpreted by analyzing the diffraction of spherical microphone array and the nonlinear distortion in measurement. In order to reduce the error introduced by the nonlinear distortion, measurement using logarithm sweep method is recommended. By setting up an appropriate measuring system, a database of spatial impulse response for far-field sound sources was established, with an azimuthal resolution of 5°(from 0° to 355°) and elevation resolution of 10°(from-50° to 90°). These data are vital to analyze the performance of the spherical microphone array capturing and binaural virtual rendering.Then, for binaural synthesis in virtual rendering, a set of head-related transfer functions(HRTF) with high spatial resolution was calculated. These far-field HRTFs with upper frequency limit of 16 k Hz and a spatial resolution of 1° were obtained by boundary element method(BEM) with head-model calibration and parameter optimization.Thirdly, an evaluation method based on binaural auditory model was proposed to analyze the spatial resolution of HRTFs needed in binaural rendering, and psychoacoustic experiments were also carried out to validate the analysis. By using the proposed method, the difference in auditory perception cause by spectral cues and interaural level difference(ILD) was distinguished, and the spatial discrimination thresholds of HRTF magnitude in binaural rendering were derived. All of these are essential for an appropriate design of binaural synthesis.Fourthly, an efficient method for dynamic binaural synthesis, which is based on principal components analysis(PCA) of HRTFs and beamforming output of the spherical microphone array, was proposed. Nine hundred virtual loudspeakers which are nearly-uniformly distributed on a spherical surface were used to rendering the signals. Multiple virtual loudspeakers was synthesized by using a PCA-based methods, and thus only 66 filters or time-domain convolutions are required in the signal processing. The efficiency of signal processing can be further improved by combining with effective dynamic interpolation method.Finally, a dynamic binaural rendering system based on spherical microphone array capturing was set up. A 64-channel spherical microphone array was used to capture the harmonic signals up to 6 order, and a dynamic binaural virtual redering with a head tracker provides the same order virtual Ambisonics reproduction. A virtual source localization experiment was also carried out to evaluate and validate the performance of system. Based on the experimental results, the influences of various factors(such as dynamic cue) on perceived performance were analyzed, which results provide helpful guidance for reference.The works in this dissertation not only evaluate and improve the performance of binaural virtual rendering of signals captured by spherical microphone array, but also provide a basis for future research.