>
> Another point : Painter explains model I, "A non linear Psychoacoustic Model
> applied to the ISO MPEG Layer 3 coder" from Baumgarte extends model I. It
> seems that a lot of publications are exploring the model I while we only
> have the model II for layer III. Did you try to adapt model I to the layer
> III? With the non-linear addition of masking thresholds, it could have good
> results too.
>
> Lionel
Okay, I'm finally starting to understand - sorry about all the
confusion :-) Here is my summary of the situation:
Painter et al. gives 3 different spreading functions:
1) approximate model with slope 25/-10
2) SF(x) = 15.81 + 7.5(x+0.474)...
3) complex formula from mpeg psy-model 1
LAME contains code for two (very similar) spreading functions,
both for psy-model 2:
4) tempx = (bval_l[i] - bval_l[j])*1.05; + other formulas
5) tempx = (bval_l[i] - bval_l[j])*(3 or 1.5) + other formulas
#4 is supposedly for layer 2, and #5 is for layer 3. #5 does not
agree with the ISO docs.
Then there is Baumgarte et al. model:
6) slope of s_l/s_u. Like the 25/-10 formula in #1, except
instead of 10, s_u is a formula which depends on the masking strength.
A couple of points:
psy-model1 vs psy-model2:
the ISO doc says that psy-model1 "can be extended to layer III" but it
is not done in the docs. Also, the Bosi et al. MPEG2-AAC paper claims
that AAC uses a version of MPEG1 psy-model 2. Finally, did Baumgarte
et al. specifiy psy-model 1? I could not find this, and assume he is
also using psy-model 2.
Finally, I think the main difference between psy-model 1 and 2 is
not the spreading function, but the identification of maskers,
and if they are "tone masking noise" or "noise masking tone".
I think either model can use any spreading function, since the
spreading function just tells how the noise masking of a given
signal extends to adjacent critical bands.
I would suggest first implementing #1 and then replacing it
with #6.
-=- MIME -=-
> CE MESSAGE EST AU FORMAT MIME. Comme votre lecteur de courrier ne comprend pas
ce format, il se peut que tout ou partie de ce message soit illisible.
--MS_Mac_OE_3023954128_235113_MIME_Part
Content-type: text/plain; charset="US-ASCII"
Content-transfer-encoding: 7bit
> When you say using -10 and 25 gives nonsense, I think you are
> confusing tempx and tempy? The values of -10 and 25 should be applied
> to tempy not tempx, as shown above. They give very reasonable results
> which agree with the curve in the Painter article: (taking x=0)
>
>
> Better than the layer 3 results, but the give much more spreading
> then is justified by the Painter article.
>
>
>
i still don't agree.:-)
The curb of the spreading function you mentioned(Figure 9b I suppose) is not
for the model II but for the model I (which has a different expression of
the spreading function) and is explained page 9, relation 28, :
-3 < dz < -1 sf =...
-1 < dz < 0 sf =...
0 < dz < 1 sf =...
1< dz < 8 sf = ...
(See step 6 of the model I on ISO docs too)You can verify that the curb of
the spreading
function is linear on each of these intervals. All the discussions of
Painter are for model I which he explains page 7.So you can not refer to
this curb as a reference.
In this case, it's OK the slopes are -10/25 around 0. And it's OK too, i
confused tempx and tempy when reading your explanations, but you mustn't
have slopes of -10/25for the model II.You can not transform the spreading
function of model II in the one of model I : they have completely different
ways of modelling the masking. This time I think Painter made a
confusion when quoting the spreading function of model II(relation(4)) with.
the slopes of model I(or maybe he didn't check) : he says on page 9 "
SF(i,j) approximates the basilar spreading(excitation pattern) described in
section II-C" which is false: he probably didn't compare both curbs.
My english may not be clear, but think about it.
Another point : Painter explains model I, "A non linear Psychoacoustic Model
applied to the ISO MPEG Layer 3 coder" from Baumgarte extends model I. It
seems that a lot of publications are exploring the model I while we only
have the model II for layer III. Did you try to adapt model I to the layer
III? With the non-linear addition of masking thresholds, it could have good
results too.
Lionel
--MS_Mac_OE_3023954128_235113_MIME_Part
Content-type: text/html; charset="US-ASCII"
Content-transfer-encoding: quoted-printable
<HTML>
<HEAD>
<TITLE>Re : [MP3 ENCODER] spreading function buggy?</TITLE>
</HEAD>
<BODY BGCOLOR=3D"#FFFFFF">
<TT><BR>
<BR>
> When you say using -10 and 25 gives nonsense, I think you are<BR>
> confusing tempx and tempy? The values of -10 and 25 should be ap=
plied<BR>
> to tempy not tempx, as shown above. They give very reasonable re=
sults<BR>
> which agree with the curve in the Painter article: (taking x=3D0)<BR>
><BR>
><BR>
> Better than the layer 3 results, but the give much more spreading<BR>
> then is justified by the Painter article.<BR>
><BR>
><BR>
><BR>
<BR>
i still don't agree.:-)<BR>
The curb of the spreading function you mentioned(Figure 9b I suppose) is no=
t<BR>
for the model II but for the model I (which has a different expression of<B=
R>
the spreading function) and is explained page 9, relation 28, :<BR>
-3 < dz < -1 sf =3D...<BR>
-1 < dz < 0 sf =3D...<BR>
0 < dz < 1 sf =3D...<BR>
1< dz < 8 sf =3D ...<BR>
(See step 6 of the model I on ISO docs too)You can verify that the curb of =
the spreading<BR>
function is linear on each of these intervals. All the discussions of<BR>
Painter are for model I which he explains page 7.So you can not refer to th=
is curb as a reference.<BR>
In this case, it's OK the slopes are -10/25 around 0. And it's OK too, i<BR=
>
confused tempx and tempy when reading your explanations, but you mustn't<BR=
>
have slopes of -10/25for the model II.You can not transform the spreading<B=
R>
function of model II in the one of model I : they have completely different=
<BR>
ways of modelling the masking. This time I think Painter made a<BR>
confusion when quoting the spreading function of model II(relation(4)) with=
.<BR>
the slopes of model I(or maybe he didn't check) : he says on page 9 " =
SF(i,j) approximates the basilar spreading(excitation pattern) described in =
section II-C" which is false: he probably didn't compare both curbs.<BR=
>
My english may not be clear, but think about it.<BR>
<BR>
Another point : Painter explains model I, "A non linear Psychoacoustic=
Model applied to the ISO MPEG Layer 3 coder" from Baumgarte extends mo=
del I. It seems that a lot of publications are exploring the model I while w=
e only have the model II for layer III. Did you try to adapt model I to the =
layer III? With the non-linear addition of masking thresholds, it could have=
good results too.<BR>
<BR>
Lionel </TT>
</BODY>
</HTML>
--MS_Mac_OE_3023954128_235113_MIME_Part--
--
MP3 ENCODER mailing list ( http://geek.rcc.se/mp3encoder/ )
--
MP3 ENCODER mailing list ( http://geek.rcc.se/mp3encoder/ )
