Front And Back Ends Signal Processing For Speech Codecs

With the development of wireless communication and networks, bandwidth resource is becoming more and more valuable. Speech communications are facing the quality problems resulting from network congestion, channel disturbance and noise, and so on. In order to reduce the rate of the speech codec and improve the utilization of bandwidth resource and the quality of the reconstructed signal, the front and back ends signal processing for speech codecs are introduced.This paper briefly introduces the principle and development of the front and back ends signal processing for speech codecs. There are four important modules in an integrated system, including Voice Activity Detector (VAD), Comfort Noise Generation (CNG), Source Controlled Rate (SCR) and Error Concealment Units (ECU).This paper focuses on the voice activity detection and comfort noise generation technologies. The differences between speech and noise are analyzed, the features of speech and noise are described, especially the principles of sub-band energy entropy and sub-band SNR. According to the research of the sub-band energy entropy and the sub-band SNR, a new VAD algorithm based on energy entropy is designed. By using the difference of the energy entropy between speech and noise, the algorithm can detect the speech and noise. In order to improve the anti-noise property of sub-band energy entropy in low SNR environment, the sub-band SNR is used to update the sub-band energy in this algorithm. Simulation and test of the proposed algorithm have shown that the performance is better than AMR and G.729B.According to the requirement of the comfort noise, a new CNG algorithm based on spectral band replication is also designed and tested in this paper. Based on the results of the simulation and test of the algorithm, the comfort noise has good effect on the silence segment as background noise. In addition the proposed algorithm reduces the bit rate of noise signal, thus reducing the bit rate of the speech codec.