Research On Voice Activity Detection Algorithm In Noise Environment

Variable rate speech coding technologies have been widely used in the terrestrial cellular mobile communication, satellite mobile communication, VoIP and other digital communication systems. Voice Activity Detection(VAD) algorithm is one of the critical technologies to realize the variable rate speech coding. Variable rate speech vocoder can use VAD algorithm to get flexible compromise between speech quality and bandwidth. Therefore, research on robust and reliable VAD algorithm has great significance in variable rate speech coding, especially in noise environment.Currently,lots of achievements have been scored in the field of VAD technology, among which the VAD algorithms based on Hidden Markov(HMM) model show better performance in distinguishing speech between background noise and thus become a key research topic of digital speech signal processing. Starting by the research background and actuality, this thesis briefly describes the basic principle of VAD algorithm and noise characteristics. Moreover, the implementation processes of VAD algorithm in Adaptive Multi-Rate(AMR) encoding standard and G.729 B encoding standard are introduced and their advantages are also analyzed. This thesis focuses on the improvement of the VAD algorithm based on HMM model and applies it to the low rate speech codec. The specific work is as follows:Firstly, in order to improve noise tracking performance of the the existing VAD algorithm based on HMM model, the Baum-Welch algorithm is used to train the noise model with noise of different characteristics and a noise library is established. During the VAD decision, the noise model in the established noise library is selected according to the different dynamic background noise. We also improves the threshold method to improve the decision more accurate. Simulation results show that the proposed algorithm has high decision accuracy and better noise tracking performance in speech signal processing.Secondly, the improved VAD algorithm is applied to the 4kb/s Mixed Excitation Linear Prediction(MELP) vocoder via Discontinuous Transmission technology to realize a variable rate speech vocoder. In the coding part of the variable rate speech encoder, speech frames are encoded by 4kb/s, while background frames are not encoded or encoder by low rate. Experimental results show that the average coding rate is significantly reduced without compromising the quality of synthesized speech, which indicates that the proposed VAD algorithm is of high practicability.