As many of you many have noticed, the net -X 3 quant code causes audible
distortion even at high bitrates. I sought out to determine how to fix
this problem.
I figured that the X3 code was giving too much importance to overall noise
and not paying enough attention to over_noise or max_noise.
After figuring out that the Absolute_E gtk theme on my roomates computer
was preventing me from seeing the noise in the GTK viewer, I got underway
testing various settings.
I found a setting I like, and it sounds MUCH better then -X3 and produces
results I feel are better then -X0. I did my testing at 160Kbit/s
but the problem shows at any bitrate. I also only tested with a few pieces
and only over small portions, because it's hard to compair when you are
eyeballing between two windows a granule at a time. :)
Here are some of my test results. Each line is the noise data from Mp3x
for each granule over three consecutive frames (group selected at random).
I'd like some feedback on how 'bad' it is to exceede the percept model, I
try to avoid it, but I'm not sure I work hard enough. I.e. will .25db over
be very like audible while -.25db wont? What kind of error threshold are
we looking at?
You can argue up and down if tot noise is more important then max noise, I
think mine sounds better, but I can't do any listening tests yet. I'm
sure it sounds better then -X3 though.
Over Max Over_n Tot_n
-X 0 (Lame 3.25 w/ def RH)
0 -0.0 0.0 -6.9
0 -0.0 0.0 -6.2
0 -0.2 0.0 -6.1
0 -0.2 0.0 -6.2
0 -0.3 0.0 -7.7
0 -0.1 0.0 -5.6
---- ---- ---- ----
0 -0.13 0.0 -6.45 < AVG of above
-X 3 (Lame 3.25 w/ def RH)
6 13.1 6.4 -9.1
6 12.8 5.5 -10.4
6 9.3 6.2 -10.0
5 7.5 5.4 -11.5
5 7.1 5.7 -12.4
5 7.9 5.4 -11.1
---- ---- ---- ----
5.5 9.61 5.76 -10.75
-X 4 (New Gregory Maxwell)
0 -0.2 0.0 -5.2
0 -0.2 0.0 -7.4
5 1.7 0.7 -8.0
4 0.6 0.3 -7.7
0 -0.5 0.0 -7.9
1 0.3 0.3 -6.9
---- ---- ---- ----
1.67 0.28 .22 -7.18
if (experimentalX==4) {
better = ( ( (0>=max_noise) && (best_max_noise>2)) ||
( (0>=max_noise) && (best_max_noise<0) && ((best_max_noise+2)>max_noise) &&
(tot_noise<best_tot_noise) ) ||
( (0>=max_noise) && (best_max_noise>0) && ((best_max_noise+2)>max_noise) &&
(tot_noise<(best_tot_noise+best_over_noise)) ) ||
( (0<max_noise) && (best_max_noise>-0.5) && ((best_max_noise+1)>max_noise) &&
((tot_noise+over_noise)<(best_tot_noise+best_over_noise)) ) ||
( (0<max_noise) && (best_max_noise>-1) && ((best_max_noise+1.5)>max_noise) &&
((tot_noise+over_noise+over_noise)<(best_tot_noise+best_over_noise+best_over_noise)) )
);
}
--
The comments and opinions expressed herein are those of the author of this
message and may not reflect the policies of the Martin County Board of
County Commissioners.
--
MP3 ENCODER mailing list ( http://geek.rcc.se/mp3encoder/ )
