Excessive end to end delay or latency in a VoIP network is one of the major drawbacks of packetized voice communication. The two main components of end to end delay are codec processing and propagation delay. Codec processing delay becomes a significant factor in centralized VoIP conferencing where voice packets, originating from the participants, are sent to a central unit or bridge to be combined using tandem mixing.
This thesis investigates a novel way based on the G.722.2 codec of mixing the packets at the central unit that reduces algorithmic complexity and therefore delay. The parameters used to represent the speech, LPCs, pitch lags, fixed codebook, and gains, are extracted from the encoded bit stream, mixed, and re-encoded instead of full decoding, mixing, and then re-encoding of the speech signals.
This parametric mixing reduces the bridge complexity by up to 85 % while still retaining acceptable speech quality, 3.7 MOS on average, as shown by simulations.