Skip to main content

Topic: Vorbis Megamix 2 based on 1.1 RC1 (Read 63416 times) previous topic - next topic

0 Members and 1 Guest are viewing this topic.
  • QuantumKnot
  • [*][*][*][*][*]
  • Developer
Vorbis Megamix 2 based on 1.1 RC1
OK, here is Vorbis Megamix 2 which is a merge of Xiph.Org 1.1 RC1, QKTune beta 3.2, and GTune 3 beta 2.  As I mentioned in the other thread, Aoyumi pointed out to me one critical value I forgot to copy over from GT3b2, hence Megamix 1 didn't have all the benefits brought about by Garf's work.

http://www.rarewares.org/quantumknot/oggenc-megamix2.exe
http://www.rarewares.org/quantumknot/vorbi...gamix2-dlls.zip (DLLs)
http://www.rarewares.org/quantumknot/vorbis-megamix2-src.zip (source code)

Changes: 

- Hopefully a more complete merge of GT3b2.
- nominal bitrate/quality is now the same as aoTuV, 1.1RC1, etc. (ie. nominal 160 kbps at -q 5, etc)
- bitrate management should work properly now

In order to aid other members to helping me find problems or things I've overlooked in the merge, I'll list the values that I've modified from the 1.1RC1 code:

In psych_44.h:

_psy_global_44
_psy_noisebias_impulse (2,3,4,5,6,7,8,9,10)

In setup_44.h:

_global_mapping_44

Things I didn't change from the 1.1RC1 code, but were different in GT3b2:

In psych_44.h:

_psy_tone_0dB
_psy_noise_suppress
_psy_ath_floater
_psy_ath_abs
_psy_stereo_modes_44
_psy_lowpass_44

In setup_44.h:

rate_mapping_44

If you believe I've missed something or that the above should reflect those in GT3b2, please let me know asap.

As always, please use this for testing.  I cannot guarantee that everything is perfect, unfortunately.
  • Last Edit: 13 July, 2004, 09:33:19 PM by QuantumKnot

  • guruboolez
  • [*][*][*][*][*]
  • Members (Donating)
Vorbis Megamix 2 based on 1.1 RC1
Reply #1
I'm probably infringing one of the holy rules of HA...
The strange "noise curtain" audible AND VISIBLE with creaking.wav is now shorten with megamix II (compared to megamix I). From 0.83 second to 0.55. Exactly like GT3b2! I can't confirm it yet by listening test, but I'm highly confident about positive quality changes on this sample.

Therefore, I think that megamix II won't suffer anymore with micro-attacks samples (which was the ultimate flaw of this encoder compared to GT3b2). I think that we have the expected federative vorbis encoder. Well done!


P.S. Of course, tests need to confirm my feeling. And I must precise than even GT3b2 wasn't fully transparent with this kind of sample. Anyway, it's a new step.
  • Last Edit: 13 July, 2004, 10:16:06 PM by guruboolez

  • indybrett
  • [*][*][*][*][*]
  • Members (Donating)
Vorbis Megamix 2 based on 1.1 RC1
Reply #2
At -q6 the average bitrate has gone up a bit. Probably more similar to GT3B2 now.

Velvet Revolver (Contraband) was average 196kbs at -q6 with Megamix I.
With Megamix II it is now 208kbs at -q6.
flac>fb2k>kernel streaming>audiophile 2496>magni>dt990 pro

  • DreamTactix291
  • [*][*][*][*][*]
  • Members (Donating)
Vorbis Megamix 2 based on 1.1 RC1
Reply #3
Right now I'm encoding an album to compare to Megamix I.
Nero AAC 1.5.1.0: -q0.45

  • Rotellian
  • [*][*]
Vorbis Megamix 2 based on 1.1 RC1
Reply #4
QK,

Ive sent you the sample.    Ive also just seen this and attemped an encode.  Whereas before the megamix @q5 and above still exhibited the same problems as below q5, megamix II does not.  It now sounds the same as the gt3b2 encode of the sample, but still very easily ABXable.    Looks like you have successfully combined the 2.  Nice job.   

I do have a question, i thought the gt3b2 tunings were only from q5 upwards, but then i found a post by Garf indicating they would be effective down to q4.    However, with gt3b2 (and megamixII now) there is a very definite change between q4.99 and q5 for this sample at least.      Is there something in particular that comes into play only at >q5?

  • QuantumKnot
  • [*][*][*][*][*]
  • Developer
Vorbis Megamix 2 based on 1.1 RC1
Reply #5
Quote
  I do have a question, i thought the gt3b2 tunings were only from q5 upwards, but then i found a post by Garf indicating they would be effective down to q4.    However, with gt3b2 (and megamixII now) there is a very definite change between q4.99 and q5 for this sample at least.      Is there something in particular that comes into play only at >q5?
[{POST_SNAPBACK}][/a]


The most obvious reason that I can think of at the moment is that it is at the boundary between two independent tunings.  Namely, my tunings at q 4 and Garf's tunings at q 5.  The region in between will be a mix of the two.

I'd be interested to see the bitrate fluctuations in this region between 4 and 5.

EDIT:  Thanks for emailing the sample.  It can be downloaded from the following URL:

[a href="http://www.rarewares.org/quantumknot/Originalsmall.flac]http://www.rarewares.org/quantumknot/Originalsmall.flac[/url]
  • Last Edit: 14 July, 2004, 06:55:36 AM by QuantumKnot

  • Rotellian
  • [*][*]
Vorbis Megamix 2 based on 1.1 RC1
Reply #6
I have just tried Garf's old GT2 encoder with interesting results for my problem sample.  Its much better than any of the current versions/tunes.

http://www.hydrogenaudio.org/forums/index....pic=23145&st=50

Vorbis Megamix 2 based on 1.1 RC1
Reply #7
wow, great!

are their binaries compiled on MSVC++ 6?
<name>madoka</name>
<uri>http://codecs.ex-sounds.net/</uri>

  • phong
  • [*][*][*][*]
Vorbis Megamix 2 based on 1.1 RC1
Reply #8
It seems like megamix 2 bitrates are quite a lot higher than other tunings -q > 4.5 or so (much higher even than gt3b2, for example).  The bitrate for megamix2 at -q 5, for instance, is almost as high as stock 1.0.1 at -q 6, and megamix 2 -q 6 is almost as high as 1.0.1 at -q 7.  Anyone wishing to do side-by-side comparisons may want to take that into account.

I'm going to update my bitrate table later on with this tuning and with the new 1.1rc1 version (which is producing bitrates different from aotuv at some -q levels).
I am *expanding!*  It is so much *squishy* to *smell* you!  *Campers* are the best!  I have *anticipation* and then what?  Better parties in *the middle* for sure.
http://www.phong.org/

  • QuantumKnot
  • [*][*][*][*][*]
  • Developer
Vorbis Megamix 2 based on 1.1 RC1
Reply #9
Quote
wow, great!

are their binaries compiled on MSVC++ 6?
[a href="index.php?act=findpost&pid=226666"][{POST_SNAPBACK}][/a]


Usually, most binaries I make are compiled on MSVC 7

  • phong
  • [*][*][*][*]
Vorbis Megamix 2 based on 1.1 RC1
Reply #10
As promised, here are equivalent -q levels.  I've had to modify the formula I use to calculate these somewhat because the variance in bitrate between the fattest vs. the leanest tuning for each bitrate is pretty big.  In a few cases an equivalent level can't be picked because of the gap between -q levels (the 2.99 to 3.0 boundry for instance):
Code: [Select]
                                                      aotuv    aotuv
-q      1.0.1   1.1rc1    aotuv     qk32    gt3b2    +qk32   +gt3b2  megamix megamix2
-1.00   -0.94    -0.58    -1.00    -0.62    -0.94    -1.00    -1.00    -1.00    -0.58
 0.00    0.00     0.07     0.02     0.26     0.00     0.02     0.02     0.02     0.07
 1.00    1.00     1.15     1.11     1.25     1.00     1.05     1.11     1.05     1.07
 2.00    n/a      n/a      n/a      2.83     n/a      2.00     n/a      2.00     2.04
 3.00    4.05     4.03     4.00     3.53     4.03     3.00     4.00     3.00     3.03
 4.00    4.69     4.65     4.63     4.32     4.36     4.00     4.33     4.00     4.02
 5.00    n/a      5.98     5.97     n/a      5.59     5.92     5.52     5.52     5.00
 6.00    6.93     6.66     6.63     6.93     6.45     6.63     6.44     6.44     6.00
 7.00    7.59     7.27     7.19     7.59     7.11     7.19     7.07     7.07     7.00
 8.00    8.64     8.44     8.41     8.64     8.48     8.41     8.31     8.31     8.00
 9.00    9.12     9.06     9.04     9.07     9.03     9.04     9.00     9.00     9.01


The graph's baseline is now 1.1rc1.
vorbis_graph.png

As always, you can get the spreadsheet used to generate these numbers here.  It has lots of other interesting gunk in it too:
OpenOffice.org
MS Excel
  • Last Edit: 13 August, 2004, 08:56:56 AM by Jan S.
I am *expanding!*  It is so much *squishy* to *smell* you!  *Campers* are the best!  I have *anticipation* and then what?  Better parties in *the middle* for sure.
http://www.phong.org/

  • QuantumKnot
  • [*][*][*][*][*]
  • Developer
Vorbis Megamix 2 based on 1.1 RC1
Reply #11
That bitrate increase at q 5 for megamix 2 is very worrying.  I'll have to look into it.

  • QuantumKnot
  • [*][*][*][*][*]
  • Developer
Vorbis Megamix 2 based on 1.1 RC1
Reply #12
Having had a look at the source again, I can't find any problems.  In order to verify this, I've made a binary of megamix 1 that includes one fix I made from GT3b2.

http://www.rarewares.org/quantumknot/oggenc-megamix-fix.exe

Phong: 

Can you do your bitrate distribution experiment comparing this fixed version of megamix 1 and megamix 2?  If they both follow similar trends, then I guess we've narrowed it down.

  • indybrett
  • [*][*][*][*][*]
  • Members (Donating)
Vorbis Megamix 2 based on 1.1 RC1
Reply #13
Quote
That bitrate increase at q 5 for megamix 2 is very worrying.  I'll have to look into it.
[a href="index.php?act=findpost&pid=227363"][{POST_SNAPBACK}][/a]


The jump at -q8 is even bigger (relatively speaking).
flac>fb2k>kernel streaming>audiophile 2496>magni>dt990 pro

  • phong
  • [*][*][*][*]
Vorbis Megamix 2 based on 1.1 RC1
Reply #14
I'd need a linux binary...
I am *expanding!*  It is so much *squishy* to *smell* you!  *Campers* are the best!  I have *anticipation* and then what?  Better parties in *the middle* for sure.
http://www.phong.org/

  • QuantumKnot
  • [*][*][*][*][*]
  • Developer
Vorbis Megamix 2 based on 1.1 RC1
Reply #15
Quote
I'd need a linux binary...
[a href="index.php?act=findpost&pid=227395"][{POST_SNAPBACK}][/a]


Oh, you only need to make a small change to the Megamix 1 source code.

In lib/modes/setup_44.h, find the static array _global_mapping_44[11] and replace the values with

Code: [Select]
1., 1., 1.5, 2., 2., 3., 3.2,  3.5, 4., 4., 4.


Then you're done.


EDIT:  Corrected value for q 4 (3. -> 2.)
  • Last Edit: 21 July, 2004, 07:47:03 PM by QuantumKnot

  • guruboolez
  • [*][*][*][*][*]
  • Members (Donating)
Vorbis Megamix 2 based on 1.1 RC1
Reply #16
I've only tested few tracks (laptop). I can't reproduce the big +17% noticed by phong at -q 5 between megamix 1 and megamix 2. Classical music is apparently less problematic. Difference is around ~5% (5..6 kbps for ~160 kbps tracks, and 10..12 kbps for ~200 kbps tracks like harpsichord).
megamix1_fixed bitrate is near megamix2. Just slightly lower (0..2 kbps).

Apparently, the changes made in lib/modes/setup_44.h are the cause (or at least one cause) of the bitrate blow of megamix2. But it's preferable to wait for the new bitrate table of phong: his values are more precise, and with more representative tracks as mine.

  • BMQ
  • [*]
Vorbis Megamix 2 based on 1.1 RC1
Reply #17
I use vorbis in movies I encode, is there significant improvement over GT3b2 at Q=6 at this point? Also, is there a libvorbis.dll version of MegaMix2 so I can try with OAGMachine?

  • QuantumKnot
  • [*][*][*][*][*]
  • Developer
Vorbis Megamix 2 based on 1.1 RC1
Reply #18
Quote
I use vorbis in movies I encode, is there significant improvement over GT3b2 at Q=6 at this point? Also, is there a libvorbis.dll version of MegaMix2 so I can try with OAGMachine?
[a href="index.php?act=findpost&pid=227679"][{POST_SNAPBACK}][/a]


For movies, I doubt you'll notice the difference between the two, unless there is a lot of classical music playing

  • guruboolez
  • [*][*][*][*][*]
  • Members (Donating)
Vorbis Megamix 2 based on 1.1 RC1
Reply #19
Even with classical... I doubt that great difference will be audible. Coarseness isn't an issue with GT3b2 beyond -q 6,00 (could be audible, but at a very little level), and therefore aoTuV can't improve a non-problematic situation.
GT3b2 is maybe better on pure attacks (castanets like), or micro-attacks parts (probably ruined by AC3). On the other side, improvements might be audible in noise/hiss area with any aoTuV-based encoder like megamix, but again, improvements are very subtle.

I'd probably use megamix rather than GT3b2, but without hope of great or major improvments.

But isn't -q 6,00 a bit high for movie encoding (DVD I suppose)? Isn't it better to keep the AC3, at least for a stereo track?
  • Last Edit: 20 July, 2004, 08:42:35 PM by guruboolez

  • BMQ
  • [*]
Vorbis Megamix 2 based on 1.1 RC1
Reply #20
Most movies I rip don't come with stereo AC3 anymore (there's probably a way to convert from 5.1 to stereo AC3 but I never looked into it).  However, from 5.1 AC3 to Vorbis -q6, I save like 1/2 the audio space. Plus you can downsample to Dolby Prologic 2 with OAGMachine which sounds pretty good on my receiver/computer speakers. Most of the time for a 2 CD rip, a Vorbis audio stream is sufficient. I still wouldn't mind if someone can compile the libvorbis.dll when they have time. I just did the Last Samurai and love listening to the endmusic on it. However I think there's some noticeable difference between the AC3 and GT3b2 vorbis version..It's instrumental stuff, lots of wind/string instrument sounds..is that like classical? 
  • Last Edit: 20 July, 2004, 09:14:15 PM by BMQ

  • john33
  • [*][*][*][*][*]
  • Developer
Vorbis Megamix 2 based on 1.1 RC1
Reply #21
I've just added oggdropXPd MegaMix2 compiles to Rarewares.
John
----------------------------------------------------------------
My compiles and utilities are at http://www.rarewares.org/

  • Thomas
  • [*]
  • Banned
Vorbis Megamix 2 based on 1.1 RC1
Reply #22
Quote
I've just added oggdropXPd MegaMix2 compiles to Rarewares.
[a href="index.php?act=findpost&pid=227766"][{POST_SNAPBACK}][/a]

Does it contains megamix 2.1? BTW. Is there any difference between 2.0 and 2.1?

  • bkvorbis
  • [*]
Vorbis Megamix 2 based on 1.1 RC1
Reply #23
Quote
I've just added oggdropXPd MegaMix2 compiles to Rarewares.
[a href="index.php?act=findpost&pid=227766"][{POST_SNAPBACK}][/a]

Can not the oggenc2.3 to compile build with the ICL7.1? 
Please.

  • john33
  • [*][*][*][*][*]
  • Developer
Vorbis Megamix 2 based on 1.1 RC1
Reply #24
Quote
Quote
I've just added oggdropXPd MegaMix2 compiles to Rarewares.
[a href="index.php?act=findpost&pid=227766"][{POST_SNAPBACK}][/a]

Can not the oggenc2.3 to compile build with the ICL7.1? 
Please.
[a href="index.php?act=findpost&pid=227783"][{POST_SNAPBACK}][/a]

Please correct me if I'm wrong, but you're asking for a ICL 7.1 MegaMix2 build of oggenc2.3, yes? I'll post one over the next few hours. I'll post when it's done.
John
----------------------------------------------------------------
My compiles and utilities are at http://www.rarewares.org/