Speech encoding and reconstruction for packet-based networks

Some speech encoding and processing algorithms for limiting speech distortion in the face of network congestion or malfunction are reviewed. These include speech waveform reconstruction, intelligent packet discarding and a low-bit-rate embedded encoding algorithm appropriate to a packet network environment. A reconstruction technique based on pitch replication extends the tolerable missing packet rate compared with the simpler approaches of silence substitution and packet repetition. The possibility of selective discarding at network nodes allows one to further increase this tolerance by an intelligent classification of speech packets according to their importance to speech quality. An elaboration of this strategy and the extension of embedded encoding techniques to the packet level permit the application of packet reconstruction schemes to low-bit-rate predictive encoding