Concealment of lost speech packets using adaptive packetization

Long-term correlation within a speech signal is usually exploited to achieve higher compression ratios (e.g. RPE-LTP coders). We aim to use the long-term correlation to influence the packetization interval of a voice stream at the sender before sending it over a lossy packet-switched network. If a packet is lost, the receiver can conceal the loss of information by using adjacent signal segments of which (due to the pre-processing/packetization at the sender) a certain similarity to the lost segment can be assumed. Subjective test results show that the adaptive packetization/concealment scheme (AP/C) can alleviate significantly the impact of isolated packet losses to speech quality