Packet　loss　and　packet　delays　due　to　packet　switching　on　IP　nets　affect　speech　quality.　This　paper　presents　an　embedded　wideband　speech　codec　which　is　one　candidate　for　the　G.VBR　proposed　by　the　telecommunication　standardization　sector　(ITU-T).　The　codec　divides　an　embedded　bit-stream　into　five　layers　with　bit　rates　from　8　kbit/s　to　32　kbit/s.　The　first　three　layers　are　coded　using　the　algebraic　code-excited　linear　prediction　(ACELP)　coding　technique　while　the　last　two　layers　are　coded　using　the　transform　coded　excitation　(TCX)　coding　technique.　When　the　current　frame　information　is　lost,　the　lost　information　is　recovered　from　classification　information　and　coded　parameters　in　the　previous　frame.　Tests　by　ITU-T　show　that　the　codec　meets　the　ITU-T　requirements.