Project Porposal - Real-time Packet loss recovery techniques for streaming audio

Packet loss can be detrimental to streaming audio causing pops and clicks. While there are several solutions to this problem, the intention is to increase the quality of reconstructed speech while keeping the cost of computation low. To attain this goal the Linear prediction based packet loss concealment method will be studied and implement. Linear prediction based algorithims extract the residual signal of the previously received speech by linear prediction analysis, uses periodic replication to generate an approximation for the excitation signal of missing speech, and generates synthesized speech using this excitation. This algorithm will be compared to other algorithms by subjective quality tests under various conditions. Shortcomings of this algorithm will be evaluated and enhancements to further improve the audio quality will be suggested.