Model the packet-loss dependent effective equipment impairment factor in speech quality estimation in VoIP and its realization

Packet loss limits the speech quality of voice over IP and the effect is described with the packet-loss dependent effective equipment impairment factor in E-model. The existing model of the factor is associated with the speech codec and limited to the network with low loss burst Ratio. With the experiments, this paper tries to find the variables in the relation with the speech quality over packet-loss network, and introduces the loss location and the distribution of the loss in burst into the model of the factor with the neural network. Compared with different models with different input parameters, the experiment results show a great prediction gain in the trained model combined with one packet loss and voice loss rate. The model can be used to predicate the speech quality in live combined with the E-model.