A Novel Error Mitigation Scheme Based on Replacement Vectors and FEC Codes for Speech Recovery in Loss-Prone Channels

In this paper, we propose an error mitigation scheme which combines two different approaches, a replacement super vector technique which provides replacements to reconstruct both the LPC coefficients and the excitation signal along bursts of lost packets, and a Forward Error Code (FEC) technique in order to minimize the error propagation after the last lost frame. Moreover, this FEC code is embedded into the bitstream in order to avoid the bitrate increment and keep the codec working in a compliant way on clean transmissions. The success of our recovery technique deeply relies on a quantization of the speech parameters (LPC coefficients and the excitation signal), especially in the case of the excitation signal where a modified version of the well-known Linde-Buzo-Gray (LBG) algorithm is applied. The performance of our proposal is evaluated over the AMR codec in terms of speech quality by using the PESQ algorithm. Our proposal achieves a noticeable improvement over the standard AMR legacy codec under adverse channel conditions without incurring neither on high computational costs or delays during the decoding stage nor consuming any additional bitrate.