If speech information is lost, it is impractical to try to retrieve it. Packet Loss Concealment (PLC) generates synthetic speech to cover the missing data.
For CELP based codecs, such as G.729, it is relatively easy since the speech is modeled during encoding phase. It is more difficult with waveform codecs, such as G.711.
However, waveform codecs can recover from packet loss more rapidly since the first speech sample in first good packet restores the speech to original.