Packet Loss Recovering Technology For Speech Transmission Over Network

The voice communication have high requirement for real-time and continuity, but now the Internet(IPv4) can only provide a Best-effort service. Network congestion will introduce the occurrence of packet loss and delay, which lead to serious decline in voice quality. In order to ensure the quality of service (QoS), some strategies need to be taken to reduce the adverse effects of voice quality caused by packet loss and delay. Voice packet loss recovery(PLR) strategy at the receiver has emerged as an important factor in determining voice quality of service.In this paper, we mainly study the voice packet loss recovery techniques at the receiver. We design and implement a series of algorithms to achieve the restoration of lost voice frames for voice quality improving on the CELP voice coding framework. Firstly we research on the framework of the CELP coding and G.723.1 speech coding, and design a packet loss model to simulate the network packet loss conditions, on which we will implement our PLR algorithms. Then we do research for the voice packet loss recovery algorithm based on time-domain waveform of the voice, as well as the parameters restoration of packet loss in CELP parameters domain using a replication and over-lap interpolation algorithm. On the basis of previous work a voicing-driven packet loss recovery algorithm for improvement is designed, which compared to the previous recovery algorithm do a better job in the voiced/unvoiced classifying approach and a more precise handling of the estimation for the transition region for the loss packet, and prove to have been better recovery effect in Signal to Noise Ratio(SNR) of the experimental results. Previous packet loss recovery algorithms largely ignore the dynamics of the statistical evolution of the speech signal, possibly leading to perceptually annoying artifacts. To address this problem, we propose the use of statistical method for packet loss recovery algorithm, no longer regarding the loss frame parameters as a determining value but a mixed Gaussian distribution GMM (Gaussian Mixture Model ), using of HMM (Hidden Markov Model) to describe the voice signal random evolution, and estimating the lost voice frame parameters by probability. Theoretically, the recovering of lost frame can have better hearing results. Finally, an objective test, the Bark Spectrum Distortion (BSD) is used to evaluate the voice quality recoverd by our algorithms. Experimental results show that the packet loss recovery algorithm based on HMM compared to replication and over-lap interpolation algorithm has better recovery effect.