An Adaptive Packet Loss Recovery Method Based on Real-time Speech Quality Assessment and Redundant Speech Transmission

In this paper, an adaptive packet loss recovery (APLR) method that improves the speech quality of a real-time speech streaming (RSS) system over IP networks is proposed. The proposed APLR method estimates the packet loss rate (PLR) of network via a real-time speech quality assessment (RSQA) at the receiver side of the RSS system, and then requests the opposite RSS system to transmit redundant speech frame data (RSD). Thus, it assists the speech decoder employed in the RSS system to reconstruct lost speech signals when the estimated PLR is high. In particular, according to the estimated PLR, the proposed APLR method then controls the bitrate of speech coding for the RSS system. In other words, a speech packet combines the bitstreams of the current speech frame data (CSD) and the RSD for a high PLR. Otherwise, for a low PLR, a speech packet consists of the CSD bitstreams alone. The effectiveness of the proposed APLR method is finally demonstrated by using an adaptive multirate-narrowband (AMR-NB) speech codec and ITU-T Recommendation P.563 as the scalable speech codec and RSQA, respectively. It is shown from experiments that an RSS system employing the proposed APLR method significantly improves the speech quality under packet loss conditions.