li xun wrote:
Dear all,
Those days I am always confused by one question that maybe is very simple in
your mind.
Here are two formulas for calculating NMR. One is NMR = Sum(abs(line^2 -
requant_line^2)) / mask_energy, the other is NMR = sum((line -
requant_line)^2)/mask_energy. Which one is correct?
In my idea, from the conversation of energy law, noise energy should be the
difference between original signal energy and re-quantized signal energy, so
the former is correct. But according to my experiment the encoding result
with the later NMR criteria is absolutely better than that with former NMR
criteria. I also review many audio codes such as mp3Lame, EAQUAL, which all
use the later NMR definition. Why? Could anyone help me to clarify this
puzzle?
You do not want the difference beween the original energy and the re-quatized
energy,
Instead you want the energy of the difference. The difference is the noise, and
the energy of the noise is a measure of how loud the noise is.
Just imagine a simple case: original = sin(x) requintized = - sin(x) (phase
inversion)
sin(x)^2 - (-sin(x))^2 = 0 two totaly different signals with equal energy: no
difference in energy
(sin(x) - (-sin(x)))^2 = 4sin(x)^2 the energy of the difference: quite
noticable.
yours
Martin
_______________________________________________
mp3encoder mailing list
[email protected]
http://minnie.tuhs.org/mailman/listinfo/mp3encoder
