An Improved Packet Loss Concealment Method for Mobile Audio Coding

In this paper, we propose an improved packet loss concealment (PLC) algorithm for ACELP-based speech coders in order to improve the quality of decoded speech at the receiver. Conventional receiver-based PLC algorithms in the AMR-WB+ speech codec are usually based on speech correlations to reconstruct the decoded speech of the lost frames by using parameter information obtained from the previous or adjacent correctly received frames. However, this approach has difficulties in reconstructing accurate voice signals when consecutive packet loss and bursts of packet loss occur. Thus, we propose a new PLC approach based on data-driven principle to recover the Immittance Spectral Frequency (ISF) coefficients of lost packet, referring to pitch period parameters of adjacent correctly received voice packets. The proposed PLC algorithm is designed for ACELP-based audio codec. We make the performance comparison between the data-driven method and the PLC algorithm currently employed in AMR-WB+ speech codec through the PESQ objective evaluation test and MOS subjective evaluation test. The results prove that our algorithm can achieve better speech recovery quality than original PLC algorithm employed in AMR-WB+ speech codec.