Research On Improvement Of VAD Reliability

Although great progresses have been made in wireless communication technology and networks technology,bandwidth of communication channels is still the valuable resource for speech communications.Speech transmission is facing the quality problems resulting from network congestion,channel disturbance,channel noise,and so on.In order to reduce the bit rate of the speech codec and improve the utilization of both bandwidth resource and quality of the reconstructed signal,the front and back ends signal processing technologies for speech codec are introduced.The voice active detection(VAD)is one of the key technologies,and it can not only maximize the save of network bandwidth,but also minimize the amount of data storage and computational complexity.The research is focused on the improvement of VAD reliability in the thesis.First,the key role and development of VAD are briefly introduced,and several common VAD and typical VAD of G.729B are studied in detail.The advantages and disadvantages of each algorithm and the reasons of decreasing SNR performances are analyzed and discussed.Second,according to the investigation of the algorithms of VAD and the consideration of the reliability,a new VAD algorithm based on sub-band energies SNR and exponential factor is designed.The improved instantaneous SNR,the long-term average SNR and the quantitative SNR of sub-band energies have shown their good robustness even in low SNR environment.The exponential factor is able to dynamically fit the envelope of the speech signal well and can be used to dynamically update the threshold and the parameters of the background noise,thus it can be applied to avoid using so many experienced values in the updating periods.Evaluation results have shown that the improved VAD algorithm designed in the thesis is able to improve the reliability of VAD in many different channel conditions.