Research On Multi-channel Acoustic Echo Cancellation Technology In Wave Field Synthesis

Wave Field Synthesis(WFS)is one of the spatial sound reproduction methods,which is based on reconstructing sound field physically to reproduce sound images.Compared with other method like Stereophony,WFS provides spatial expression without sweet-pot,which means sound can be perceived correctly in effective zone.In recent years,the development of network and telecommunication makes it possible to utilize immersive sound reproduction system such as WFS in full-duplex communication.Meanwhile,Acoustic Echo Cancellation(AEC)is necessary in full-duplex communication.Therefore,the key to combining WFS with full-duplex communication like video conferencing is to solve the acoustic echo problem caused by the usage of loudspeaker array in WFS efficiently and effectively.The work of this thesis is focused on reducing the computational complexity of multi-channel acoustic echo canceller and improving its effectiveness and robustness in the application of immersive video conferencing.The main work of this thesis is introduced as follows.Firstly,the traditional multi-channel AEC framework is modified by taking the object-based characteristics of WFS into consideration.In this work,sound objects are regarded as far-end signals in echo canceller instead of loudspeaker's signals in traditional AEC,leading to the great decrease in computational complexity and the avoidance of the degraded performance caused by the coherence between channels.In addition,the partitioned frequency domain fast block adaptive method is taken into the framework to solve the lengthy adaptive filter led by the usage of sound objects as far-end.Combining the prementioned two steps,we obtain an efficient frequency domain framework of multichannel acoustic echo canceller in WFS scenario.Finally,several experiments are conducted and the results validate the efficiency and effectiveness of the above framework.Secondly,several improving methods on the aforementioned efficient framework of multi-channel acoustic echo cancellation in specific immersive video conferencing based on WFS are proposed.Considering the possible impulsive disturbance or noise,a spectral entropy based voice activity detection method,a deep neural network based keyboard impulse detection module,and a noise suppression based on spectrum subtraction method are fused into the above AEC framework to enhance the effectiveness and robustness.To validate the improved effectiveness and robustness,experiments based on recorded data are conducted and results prove that the joint framework perform better in forms of effectiveness and robustness.