I have been working on building up a acoustic echo
canceller on a digital phone based on PCM.
With many reference I have found with this group
on double-talk functionality and near-end noise
cancellation, I have discovered that a more robust
system will be inverse-propertionate to the rate
of initial convergence.
Where does this result from? With faster and faster
DSPs on the market, I would assume a simple adaptation
process would not affect that much to the processing
speed. Yet, the phone still seems to have a slower