Hi all,

I am working on psychoacoustic model II as explained in ISO Reference 14496-3 Sub-part 4. I have been reading about the discrepancies in 14496-3 and I feel even the clarification of those discrepancies are somewhat in gray. So I want to ask, not a new question, but probably in a different way.

From my experiments, I have observed that the window switching decision is a bit slow in switching back to long window from short window.

For the following time signal:
TimeSignal = zeros(1,1024*3.5);
TimeSignal(1024*3.5+1:1024*4.5) = sin(2*pi*10000*t(1:1024*1));
TimeSignal(length(TimeSignal)+1:2*length(TimeSignal))= TimeSignal;
TimeSignal(length(TimeSignal)+1:2*length(TimeSignal))= TimeSignal;


I have observed that the short window is triggered with the first attack (i.e. instantly). So it seems that long to short window switching is fine.

But when the attack is removed, the short to long switch is not instant. It takes one sample block to switch back to long window.

The equation of perceptual entropy in 14496-3 says,

perceptual_entropy = perceptual_entropy - ((w_high(b) - w_low(b)) * log10(part_energy_thr(b)/(part_energy(b)+1)));

The above equation states that the window switching decision is a function of partition energy and 'pre echo control & Threshold in quiet' calculations on actual energy threshold for each partition.

It appears that the affect of previous sample(s) is more on perceptual entropy than it should be!!!


My Question is: What are the reasons for delay in switching from short to long window?

Regards,
Nabsha

Much better psycho model you will find in 3GPP TS 26.403.