A speech packet loss concealment algorithm using real-time speech quality measurement and redundancy coding

Abstract

In this paper, a new speech packet loss concealment (PLC) algorithm is proposed for improving speech quality within real-time wideband speech streaming systems over IP networks. In order to guarantee seamless speech quality within wideband speech streaming systems, the proposed PLC algorithm first estimates a packet loss rate (PLR) and then requests a sender to transmit redundant speech data when estimated PLR is assumed to be high. In particular, PLR is estimated by utilizing real-time speech quality measurement at the receiver side of the streaming system. Such PLR estimation is effective in two aspects. First, speech quality under real-time streaming is mostly correlated with PLR. Second, speech quality is measured as a mean opinion score (MOS) which is considered a clearer indicator of speech quality than any other parameter. Based on the estimated PLR, the proposed PLC algorithm controls the bit-rate of a wideband speech codec. That is, when the PLR is estimated to be high, a speech packet of the current frame combines the streams of the current and future frames in order to assist a speech decoder to reconstruct speech signals from the packet of the previous frame. On the other hand, when a low PLR is estimated, the speech packet consists of the current frame streams alone. The effectiveness of the proposed PLC algorithm is demonstrated by using the adaptive multi rate-wideband (AMR-WB) codec and ITU-T Recommendation P.563 as a wideband speech codec and an objective speech quality measurement method, respectively. It is shown from the experiments that a wideband speech streaming system employing the proposed PLC algorithm significantly improves wideband speech quality under packet loss conditions.