A nonlinear prediction speech coding ADPCM algorithm based on RNN

In this paper, the nonlinear prediction speech coding ADPCM algorithm based on the recurrent neural network (RNN) is introduced. In this coding system, the bit rate is not increased by the backward predictor and the long-term correlation is extracted because of the inner memory of the RNN. The experiment results show that the speech coding system based on RNN has a lower dimension and less hidden layers than that of BP network but without a decrease in the quality of speech. Besides, the speech quality of the new system is better than that of the standard G.721 speech coding system.