Markov chain prediction for missing speech frame compensation

Transmitting voice over packet-switched networks, such as the Internet, is an appealing communication alternative to the traditional wireline system. The ability to lower the cost of long-distance telephone calls and provide additional capabilities is attracting customers worldwide to this tool. However, many current packet-switched protocols cannot guarantee real-time delivery of packets. When voice packets are lost, deleted, or excessively delayed in the network, the receiver must provide something for the listener to hear. This paper describes Markov chain prediction, a technique for compensating when speech frames are missing. It outperforms venerable frame repetition using both subjective and objective measurements.