In the ISO AAC specs, there is pre-echo control applied to the calculated masking energy..

nb = min ( nb, scale*prev_nb)

where scale = 2 for long block
scale = 1 for short block

In the case of long blocks, I was wondering why it is still necessary to "bound" the current masking energy with that of the previous frame since the unpredictibility measurement is already taken as the minimum of the 8 short blocks ???

Has this anything to do with the fact, pre-echo can still be caused by "slow" energy changes across 2 long blocks ?? and the failure of the 8 short blocks to produce a sufficiently low enough unpredictibility measurement ??

It seemed that this technique would result it very high bitrates..

wkwai

I would say the ISO-method for detecting short-blocks is useless, as are the other non-normative proposed encoding methids in the AAC standard document.

/Pontus

Do you mean the fft switching technique for block switching ?
There is certainly some limitation there.. especially if the attacks are too closely spaced as in the case of fatboy..

But I am interested in cases where there are small temporal changes in time domain but isn't enough to cause the need to switch to short blocks..
Compensation can be made on the psychoacoustic model and the application of TNS / Gain control on the mdct spectrals..

wkwai