IPB

Welcome Guest ( Log In | Register )

 
Reply to this topicStart new topic
lame --ap-s bitrate goes wild, on loud tracks
sony666
post Apr 2 2003, 13:43
Post #1





Group: Members
Posts: 573
Joined: 22-February 02
Member No.: 1375



Hi all,
when playing around with wavegain, I made a rather surprising experience.
I grabbed a track of a very "hot" (pushed to max volume) CD and encoded it once normally and once after applying wavegain. both times with dibrom's lame 3.90.2 and --ap-s, no other switches.

lame output of the "normal" encoding (the original wav file):

lame --alt-preset standard "Warlock - Time to Die.wav" orig_aps.mp3
LAME version 3.90.2 MMX (http://www.mp3dev.org/)
-- Compiled at http://www.hydrogenaudio.org

CPU features: i387, MMX (ASM used), SIMD, SIMD2
Using polyphase lowpass filter, transition band: 18671 Hz - 19205 Hz
Encoding Warlock - Time to Die.wav
to orig_aps.mp3
Encoding as 44.1 kHz VBR(q=2) j-stereo MPEG-1 Layer III (ca. 7.4x) qval=2
32 [ 2] %
128 [ 20] %
160 [ 337] %%*****
192 [ 1630] %%%%%%%%%************************
224 [ 3341] %%%%%%%%%%%%%%%%%%%%%%********************************************
256 [ 3287] %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%**************************
320 [ 1641] %%%%%%%%%%%%%%%%%%%%%%%%%%*******
average: 242.2 kbps LR: 4859 (47.37%) MS: 5399 (52.63%)


now the lame result of the wavegained input:

lame --alt-preset standard wavgain_std.wav wavgained.mp3
LAME version 3.90.2 MMX (http://www.mp3dev.org/)
-- Compiled at http://www.hydrogenaudio.org

CPU features: i387, MMX (ASM used), SIMD, SIMD2
Using polyphase lowpass filter, transition band: 18671 Hz - 19205 Hz
Encoding wavgain_std.wav
to wavgained.mp3
Encoding as 44.1 kHz VBR(q=2) j-stereo MPEG-1 Layer III (ca. 7.4x) qval=2
32 [ 5] %
128 [ 52] %
160 [ 1090] %%%%%%%********
192 [ 4908] %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%*********************************
224 [ 3325] %%%%%%%%%%%%%%%%%%%%*************************
256 [ 765] %%%%%%%****
320 [ 113] %*
average: 204.8 kbps LR: 4860 (47.38%) MS: 5398 (52.62%)


to make it complete, the output of wavegain.exe (0.9.8 win32) for the second encoded wav:

Gain | Peak | Scale | New Peak | Track
---------------------------------------------------------
-8.42 dB | 32549 | 0.38 | 12346 | wavgain_std.wav


So basically the average bitrate for the original wav is 242.2k, the bitrate of the wavgained encode 204.8k.
I mean, I expected a small difference, but 37,4kbit/s a VERY hefty imho when you consider that both mp3 sound exactly the same after RG scanning them in fb2k.

Should wavegaining before encoding be recommended to reduce bitrates for a large portion of today's music? Dang, I wonder how much diskspace is wasted here on all my "hot" Hard&Heavy tracks I didnt wavegain blink.gif

This post has been edited by sony666: Apr 2 2003, 13:54
Go to the top of the page
+Quote Post
sony666
post Apr 2 2003, 14:10
Post #2





Group: Members
Posts: 573
Joined: 22-February 02
Member No.: 1375



--a-p extreme does not differ so much in bitrate, but look at the Joint Stereo handling:

normal:

32 [ 2] %
128 [ 8] %
160 [ 141] %**
192 [ 895] %%*************
224 [ 2224] %%%%%%********************************
256 [ 3036] %%%%%%%%%%%%%%%%%%%%%%%****************************
320 [ 3952] %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%*************
average: 266.7 kbps LR: 4952 (48.27%) MS: 5306 (51.73%)


wavegained:

32 [ 4] %
128 [ 21] %
160 [ 156] %**
192 [ 800] %%%%%%*******
224 [ 2671] %%%%%%%%%%%%%%%%%%%%%%%%%%%%***************
256 [ 4109] %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%************
320 [ 2497] %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%****
average: 256.4 kbps LR: 7699 (75.05%) MS: 2559 (24.95%)

The wavegained encode has 75% "true stereo" blocks against 48% on the normal wav file. Very strange

This post has been edited by sony666: Apr 2 2003, 14:11
Go to the top of the page
+Quote Post
JohnV
post Apr 2 2003, 15:01
Post #3





Group: Developer
Posts: 2797
Joined: 22-September 01
Member No.: 6



Try the same test, but using also -Y switch. It may be that one reason to this big difference with APS is the sfb21 problem of vbr mp3.


--------------------
Juha Laaksonheimo
Go to the top of the page
+Quote Post
tigre
post Apr 2 2003, 15:35
Post #4


Moderator


Group: Members
Posts: 1434
Joined: 26-November 02
Member No.: 3890



This test shows similar things. So a possible reason could be that by applying wavegain some parts of the signals are moved below the encoder's ATH, so it decides there's no need to encode them.


--------------------
Let's suppose that rain washes out a picnic. Who is feeling negative? The rain? Or YOU? What's causing the negative feeling? The rain or your reaction? - Anthony De Mello
Go to the top of the page
+Quote Post
sony666
post Apr 2 2003, 15:39
Post #5





Group: Members
Posts: 573
Joined: 22-February 02
Member No.: 1375



Thanks for your answer John smile.gif

-lame 3.90.2 --alt-preset standard -Y (on the original wav rip):

*** WARNING *** the meaning of the experimental -Y has changed!
now it tells LAME to ignore sfb21 noise shaping (VBR)
LAME version 3.90.2 MMX (http://www.mp3dev.org/)
Using polyphase lowpass filter, transition band: 18671 Hz - 19205 Hz
Encoding as 44.1 kHz VBR(q=2) j-stereo MPEG-1 Layer III (ca. 7.4x) qval=2
32 [ 2] %
128 [ 141] %*
160 [ 4782] %%%%%%%%%%%%%%%%%%%%******************************************
192 [ 5130] %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%************************
224 [ 106] %*
256 [ 22] %
320 [ 75] %
average: 177.6 kbps LR: 4859 (47.37%) MS: 5399 (52.63%)

*************

-lame 3.90.2 --alt-preset standard -Y (on the wavegained wav):

Using polyphase lowpass filter, transition band: 18671 Hz - 19205 Hz
32 [ 5] %
128 [ 172] %**
160 [ 4584] %%%%%%%%%%%%%%%%%%%****************************************
192 [ 5212] %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%*************************
224 [ 181] %%*
256 [ 27] %
320 [ 77] %
average: 178.2 kbps LR: 4860 (47.38%) MS: 5398 (52.62%)

*************

-lame 3.90.2 --alt-preset extreme -Y (on the original rip):

Using polyphase lowpass filter, transition band: 19383 Hz - 19916 Hz
32 [ 2] %
128 [ 135] %*
160 [ 4609] %%%%%%%%%%%%%%%%%%%%***************************************
192 [ 5205] %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%************************
224 [ 170] %%*
256 [ 52] %
320 [ 85] %*
average: 178.7 kbps LR: 4952 (48.27%) MS: 5306 (51.73%)

*************

-lame 3.90.2 --alt-preset extreme -Y (on the wavegained wav):

Using polyphase lowpass filter, transition band: 19383 Hz - 19916 Hz
32 [ 4] %
128 [ 82] %*
160 [ 2409] %%%%%%%%%%%%%*******************
192 [ 5035] %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%*************
224 [ 1858] %%%%%%%%%%%%%%%%%%%%%%%%*
256 [ 664] %%%%%%%%%
320 [ 206] %%%
average: 196.4 kbps LR: 7699 (75.05%) MS: 2559 (24.95%)

*************

Hmm.. that looks a lot more reasonable for --a-p s, but --a-p e is still acting up on the wavegained file (JS frames, 18k higher bitrate than to original)
Now I dont know what to make of this... doesnt -Y cut off at 16Khz (thats what I read here a lot) or is the encoder-output lowpass correct?
Whats the impact on quality when bitrate differs a whole 65k for the aps / aps -Y files... huh.gif

This post has been edited by sony666: Apr 2 2003, 15:41
Go to the top of the page
+Quote Post
JohnV
post Apr 2 2003, 16:50
Post #6





Group: Developer
Posts: 2797
Joined: 22-September 01
Member No.: 6



With APS the situation is clear. At first there were louder high frequencies which according to psychoacoutics needed more resolution, but because of SFB21 problem of mp3, the bitrate bloats with vbr if the psyac thinks it needs to give very high resolution to over 16khz frequencies. The result of lack of scalefactor for ScaleFactorBand21 in that case (sfb21 covers the over 16khz frequencies) is that lower than 16khz frequencies will be encoded with unnecessarely high resolution (as funny as it sounds), thus the bloat!

When you applied wavegain, the higher frequencies didn't need that kind of resolution according to psychoacoustics (because those were probably less relatively audible/lower energy high freqs), thus it didn't result that kind of bloat.

When you applied APS -Y, it disabled noise shaping for over 16khz frequencies (sfb21), and the bloating effect disappers, but only some of the strongest high frequencies will be encoded.

With --alt-preset extreme -Y the bitrate difference obviously results from the higher use of stereo frames with the wavegained original. Extreme preset gives more easily stereo frames than APS. The reason why the lr/ms frame distribution changed is that some of the masking properties changed when you applied wavegain, and since the decicion of using lr/ms is based on masking calculations, and since extreme gives more easily stereo frames, the distribution changed quite a lot, and the bitrate increased.


--------------------
Juha Laaksonheimo
Go to the top of the page
+Quote Post
Lev
post Apr 2 2003, 16:54
Post #7





Group: Members
Posts: 524
Joined: 7-November 02
From: Gloucester, UK
Member No.: 3716



ahhh

- thanks, I followed that thread with great interest


--------------------
http://www.megalev.co.uk
Go to the top of the page
+Quote Post
sony666
post Apr 2 2003, 16:56
Post #8





Group: Members
Posts: 573
Joined: 22-February 02
Member No.: 1375



why not use 3.94 b12 also smile.gif

-lame 3.94 b12 --alt-preset standard (on the original wav):

LAME version 3.94 MMX (alpha 12, Mar 31 2003 12:01:50) (http://www.mp3dev.org/)
warning: alpha versions should be used for testing only
CPU features: i387, MMX (ASM used), SIMD, SIMD2
Using polyphase lowpass filter, transition band: 18671 Hz - 19205 Hz
Encoding orig.wav to 394b12_orig_aps.mp3
Encoding as 44.1 kHz VBR(q=4) j-stereo MPEG-1 Layer III (ca. 10x) qval=3
32 [ 2] *
96 [ 8] *
112 [ 40] %
128 [ 162] %**
160 [ 1588] %%************************
192 [ 4114] %%%%%%%%%%%%%%%%%*************************************************
224 [ 3383] %%%%%%%%%%%%%%%%%%%%%%%%%******************************
256 [ 843] %%%%%%********
320 [ 118] %%
average: 202.9 kbps LR: 3022 (29.46%) MS: 7236 (70.54%)

*********

-lame 3.94 b12 --alt-preset standard (on the wavegained sample):

Using polyphase lowpass filter, transition band: 18671 Hz - 19205 Hz
Encoding as 44.1 kHz VBR(q=4) j-stereo MPEG-1 Layer III (ca. 10x) qval=3
32 [ 5] %
96 [ 20] %
112 [ 46] %
128 [ 215] %**
160 [ 2574] %%%***************************
192 [ 5824] %%%%%%%%%%%%%%%%%%%%%%%%******************************************
224 [ 1303] %%%%%%*********
256 [ 191] %%*
320 [ 80] %
average: 188.3 kbps LR: 3025 (29.49%) MS: 7233 (70.51%)

***********

-lame 3.94 b12 --alt-preset standard -Y (on the original wav):

32 [ 2] *
96 [ 56] *
112 [ 277] %****
128 [ 781] %************
160 [ 3814] %%%%%%%%%****************************************************
192 [ 4174] %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%***********************************
224 [ 907] %%%%%%%********
256 [ 166] %%*
320 [ 81] %*
average: 177.4 kbps LR: 3022 (29.46%) MS: 7236 (70.54%)

*************

-lame 3.94 b12 --alt-preset standard -Y (on the wavegained sample):

32 [ 5] %
96 [ 75] %*
112 [ 277] %****
128 [ 814] %************
160 [ 3694] %%%%%%%%%*************************************************
192 [ 4228] %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%************************************
224 [ 927] %%%%%%%********
256 [ 163] %%*
320 [ 75] %*
average: 177.3 kbps LR: 3025 (29.49%) MS: 7233 (70.51%)

*********
*********

-lame 3.94 b12 --alt-preset extreme (on the original wav):

Using polyphase lowpass filter, transition band: 19383 Hz - 19916 Hz
Encoding as 44.1 kHz VBR(q=4) j-stereo MPEG-1 Layer III (ca. 10x) qval=3
32 [ 2] %
128 [ 12] %
160 [ 51] %
192 [ 446] %%%%%%**
224 [ 3088] %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%**
256 [ 3868] %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%*
320 [ 2791] %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%*
average: 260.3 kbps LR: 9855 (96.07%) MS: 403 (3.929%)

*********

-lame 3.94 b12 --alt-preset extreme (on the wavegained sample):

32 [ 3] %
128 [ 22] %
160 [ 54] %
192 [ 655] %%%%%%%%%**
224 [ 4115] %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%**
256 [ 4248] %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%*
320 [ 1161] %%%%%%%%%%%%%%%%%%*
average: 245.5 kbps LR: 9854 (96.06%) MS: 404 (3.938%)

*******

-lame 3.94 b12 --alt-preset extreme -Y (on the original wav):

32 [ 2] %
128 [ 32] %
160 [ 1094] %%%%%%%%%%%*
192 [ 6156] %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%**
224 [ 2157] %%%%%%%%%%%%%%%%%%%%%%%*
256 [ 586] %%%%%%%
320 [ 231] %%%
average: 201.6 kbps LR: 9855 (96.07%) MS: 403 (3.929%)

*********

-lame 3.94 b12 --alt-preset extreme -Y (on the wavegained sample):

32 [ 3] %
128 [ 38] %
160 [ 1090] %%%%%%%%%%%*
192 [ 6165] %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%**
224 [ 2165] %%%%%%%%%%%%%%%%%%%%%%%*
256 [ 569] %%%%%%*
320 [ 228] %%%
average: 201.5 kbps LR: 9854 (96.06%) MS: 404 (3.938%)

******

phew, sorry for spamming the forum smile.gif My n00biish observations for 3.94 b12:
--ap extreme looks VERY aggressive on the LR-Stereo usage in 3.94 b12, no matter if gained and/or use of -Y
- the -Y switch gives steady and for my imagination, reasonable bitrates

This post has been edited by sony666: Apr 2 2003, 17:10
Go to the top of the page
+Quote Post
sony666
post Apr 2 2003, 19:01
Post #9





Group: Members
Posts: 573
Joined: 22-February 02
Member No.: 1375



QUOTE (JohnV @ Apr 2 2003 - 04:50 PM)
With APS the situation is clear. At first there were louder high frequencies which according to psychoacoutics needed more resolution, but because of SFB21 problem of mp3, the bitrate bloats with vbr if the psyac thinks it needs to give very high resolution to over 16khz frequencies. The result of lack of scalefactor for ScaleFactorBand21 in that case (sfb21 covers the over 16khz frequencies) is that lower than 16khz frequencies will be encoded with unnecessarely high resolution (as funny as it sounds), thus the bloat!

When you applied wavegain, the higher frequencies didn't need that kind of resolution according to psychoacoustics (because those were probably less relatively audible/lower energy high freqs), thus it didn't result that kind of bloat.

When you applied APS -Y, it disabled noise shaping for over 16khz frequencies (sfb21), and the bloating effect disappers, but only some of the strongest high frequencies will be encoded.

With --alt-preset extreme -Y the bitrate difference obviously results from the higher use of stereo frames with the wavegained original. Extreme preset gives more easily stereo frames than APS. The reason why the lr/ms frame distribution changed is that some of the masking properties changed when you applied wavegain, and since the decicion of using lr/ms is based on masking calculations, and since extreme gives more easily stereo frames, the distribution changed quite a lot, and the bitrate increased.

most interesting smile.gif
Might there be any chance that "the stuff -Y does" (or something very similar with a bit less impact but still stabilizing bitrates) will be defaulted in 3.94 for -a-p's?

It's just like, this was not a very difficult sample imho (standard Heavy Metal), I stumbled upon the bitrate discrepancy just by accident while testing wavegain. -Y reduced bitrate by 65k for aps/3.90.2, 88k (jeez) for ape/3.90.2, 25k for aps/3.94b12, 58k for ape/3.94b12 (the wavegain stuff not considered)

This post has been edited by sony666: Apr 2 2003, 19:09
Go to the top of the page
+Quote Post
Gabriel
post Apr 3 2003, 10:01
Post #10


LAME developer


Group: Developer
Posts: 2950
Joined: 1-October 01
From: Nanterre, France
Member No.: 138



Heavy metal is considered to be difficult for mp3.
Go to the top of the page
+Quote Post
JohnV
post Apr 3 2003, 11:06
Post #11





Group: Developer
Posts: 2797
Joined: 22-September 01
Member No.: 6



QUOTE (sony666 @ Apr 2 2003 - 09:01 PM)
Might  there be any chance that "the stuff -Y does" (or something very similar with a bit less impact but still stabilizing bitrates) will be defaulted in 3.94 for -a-p's?

Well.. if you increase masking (decrease resolution) of sfb21 frequencies, then it prevents the bloat (prevents unnecessarely high resolution of under sfb21 frequencies).
APS already does this by using --ns-sfb21 3.75 but of course if you increase the over 16khz masking, it will result less high frequencies. As said before this whole thing is a problem in mp3 technical specs. If MP3 had a scalefactor for adjusting sfb21 resolution, this wouldn't be an issue.
I think Dibrom pretty much found out that increasing sfb21 masking by 3.75dB is quite optimal considering both quality and bloat, so the same figure is used in alpha builds. May be it's worth checking again because 3.94alphas have a bit different masking properties than stable builds.

Heavy metal is pretty much the type of music which results often this bloat effect.


--------------------
Juha Laaksonheimo
Go to the top of the page
+Quote Post

Reply to this topicStart new topic
1 User(s) are reading this topic (1 Guests and 0 Anonymous Users)
0 Members:

 



RSS Lo-Fi Version Time is now: 20th December 2014 - 16:52