Research On VAD With Optimized Algorithms

With the fast development of wireless communication and networks, bandwidthresource of communication channels is becoming more and more valuable. Speechcommunications are facing the quality problems resulting from network congestion,channel disturbance, noise and so on. For the purpose of reducing the bit rate of thespeech codec, increasing the efficiency of channel bandwidth utilization and improvingthe voice quality of speech communication, the voice activity detection (VAD) forspeech signal processing is introduced.In order to reduce the calculation complexity in the existing VAD algorithms forthe application of hardware platforms with limited processing abilities, the followingtasks have been done in the thesis: First, theory and application of VAD have beenstudied, then the VAD algorithms and their performances were analyzed and tested;Secondly, the VAD algorithms of3GPP AMR has been evaluated in details; Thirdly,the possibility of reducing the algorithm complexity of VAD in AMR codec has beenstudied and the lower complexity VAD algorithms have been proposed; Finally, theperformance of the optimized algorithms have been tested.Test results have shown that the optimized VAD algorithms is better than theoriginal method on the detection time, and the detection rate of the optimizedalgorithms is slightly poorer with clean speech and higher SNR in30dB,20dB and10dB; while with0dB SNR, the performance of the optimized algorithm falls rapidlycompared with the original algorithms. The objective PESQMOS score also shows thatquality of the synthesized speech processed by the optimized VAD algorithms is poorerthan the original algorithms.