New AAC encoder(falabaac), Best quality, 128kbps support 22kHz, 96kbps also support 20kHz bandwidh for 44.1kHz sample rate Options

[blueheart73]

URL for download source code and product:
http://code.google.com/p/falab/downloads/list


This is the new aac encoder , different from FAAC, I think it is better than FAAC.
And I compare with the qtaac and neroAAC, I also think falabaac is better than both of them when bitrate >= 96kbps(now falabaac only support LC, HEv2 is developing)

quality:
128kbps hear almost lossless(support 22kHz bandwidth), and 96kbps is very good(support 20kHz bandwidth)

speed:
6 speed for your choice, default is -l 2, is fast enough, and use -l 4 (faster than current all aac encoder)

other:
open source code, GNU license

download them from:
http://code.google.com/p/falab/downloads/list

please download the latest version: falabaac-v2.0.0.178.tar.gz

Need all of you test, and if have any questions and bugs please email to me : luolongzhi@gmail.com

[alter4]

...I also think falabaac is better than both of them when bitrate >= 96kbps(now falabaac only support LC, HEv2 is developing)

Daring statement! I'll try to ABX some samples at 128kbps.

[greynol]

...a statement that should not have been made without DBT data.

That the OP is specifying frequency response causes me to wonder if he knows how to assess quality properly.

This post has been edited by greynol: Jan 1 2013, 17:45

[lvqcl]

I compiled the encoder with MSVC2010, but with suspicious warnings:

CODE

falabaac\libfalabaac\fa_aacquant.c(1234): warning C4700: uninitialized local variable 'gr' used
falabaac\libfalabaac\fa_tns.c(215): warning C4700: uninitialized local variable 'k' used
falabaac\libfalabaac\fa_mdctquant.c(907): warning C4700: uninitialized local variable 'enrgs' used
falabaac\libfalabaac\fa_mdctquant.c(908): warning C4700: uninitialized local variable 'maxs' used
falabaac\libfalabaac\fa_mdctquant.c(910): warning C4700: uninitialized local variable 'enrgd' used
falabaac\libfalabaac\fa_mdctquant.c(911): warning C4700: uninitialized local variable 'maxd' used
falabaac\libfalabaac\fa_mdctquant.c(913): warning C4700: uninitialized local variable 'enrgl' used
falabaac\libfalabaac\fa_mdctquant.c(914): warning C4700: uninitialized local variable 'enrgr' used
falabaac\libfalabaac\fa_mdctquant.c(916): warning C4700: uninitialized local variable 'maxl' used
falabaac\libfalabaac\fa_mdctquant.c(917): warning C4700: uninitialized local variable 'maxr' used

[db1989]

...a statement that should not have been made without DBT data.

That the OP is specifying frequency response causes me to wonder if he knows how to assess quality properly.

Seriously this. Has lossy encoding reached a point where it’s not absurd to (attempt to) encode 22.05 kHz at a mere 128 kbps? I doubt it.

Edit to clarify, just in case: I was referring to a sampling rate of 44.1 kHz with no lowpass, i.e. 22.05 kHz as the maximal frequency (well, just over it, technically), not the sampling rate.

This post has been edited by db1989: Jan 2 2013, 00:27

[sluggy]

Has anyone managed to test this out yet?

[IgorC]

To begin with, as we all know that the human upper limit of hearing is 20 kHz.
One could argue that some 7 years old kids can hear 21 kHz pure tone but they rather won't appreciate or even notice that in real case . Anyway it's entirely another discussion.


An excellent and very mature LC-AAC encoders as Apple and FhG can preserve approx. 15.7-17kHz at 96-128 kbps respectively. Given that AAC (LC profile) format is mature and actually old it's very unlikely (impossible) to get any better numbers.
That's all staying on topic.

Now it's a whole different story if a format can use bandwidth extension. For example HE-AAC can achieve 20 kHz at 64 kbps. Of course, SBR wasn't design to be transparent but since a new bandwidth extensions have appeared (if somebody is interested google for a literature).

Opus has a high grade of transparency at 128 kbps (VBR) and use a bandwidth extension which preserves 20 kHz bandwidth. Both Opus and LC-AAC are on par at 128 kbps (VBR) and it would be difficult to see which approach is better (AAC's lowpass at 17 kHz or Opus high quality bandwidth extension up to 20 kHz).

Speaking of my experience, I can ABX 16 kHz tone and still can do it with 17 kHz but I don't hear the last one rather feel it. 18kHz isn't perceived by me in any way.

So when I hear a lossy file with enough high bitrate and 16 kHz bandwidth I can't spot any artifacts but feel that something is different. It's still ABXable but scores are very high like 4.8-4.9 of max 5.0.
It's ABXable because of absence of energy from high frequencies as I don't actually enable to hear a details in that range.
That's where bandwidth extension makes a good job. So my humble guess is that a good bandwidth extension can be superior to a classic lowpass at transparent rates as it can preserve an energy at extremely low bitrate cost.

This post has been edited by IgorC: Jan 1 2013, 21:11

[AiZ]

Hello everybody and happy new year,

Has anyone managed to test this out yet?

Have just tried the Windows binary provided in falabaac-v2.0.0.178.tar.gz, I can say it is not that transparent at 128kbit/s :

CODE

foo_abx 1.3.4 report
foobar2000 v1.1.18
2013/01/01 23:39:46

File A: C:\Program Files (x86)\qaac\cmm.flac
File B: D:\temp\falabaac\cmm_128.aac

23:39:46 : Test started.
23:41:23 : 01/01  50.0%
23:41:31 : 02/02  25.0%
23:41:40 : 03/03  12.5%
23:41:47 : 04/04  6.3%
23:41:55 : 05/05  3.1%
23:42:04 : 06/06  1.6%
23:42:11 : 07/07  0.8%
23:42:19 : 08/08  0.4%
23:42:25 : 09/09  0.2%
23:42:31 : 10/10  0.1%
23:42:35 : Test finished.

----------
Total: 10/10 (0.1%)

And I've not taken the most difficult song to encode (Carly Rae Jepsen - Call me maybe)...

Bye,


AiZ

[blueheart73]

Thanks for all of your advice and attentions!!
The purpose of this encoder is to supply a totally free software which has good encoding quality of AAC, maybe the quality test is not enough now because of the short time development.
In my experiments, I use the sqam which supply from the EBU test audio files and some music samples I have. I optimize the encoder serverly times(espically the quantize method, the key factor which influence the sound quality, I use 3 method and found that the last one is good and I satisfy with it, but maybe is also not good enough).
So , I need you guys to help me to test the encoder problems and bugs, and tell me what the problems is ( eg: frequency response problem, or for same special sample performance is worse), I hope you can send me the problems and the problem audio sample is better!

If my opinions is not right , pardon me and help me, I think this encoder will be bettern if I find what the problem with it!!
And I really want to know the difference compare with the qtaac and neroaac, I hope you can help me to do the test experiment and give me the results(I want to know where is the faults or problems, and I can fix it and optimize it )

Thanks for your attentions and happy new year!!

[LithosZA]

I can say it is not that transparent at 128kbit/s...

FAAC also isn't transparent at 128Kbit/s. How does falabaac compare to FAAC?

[greynol]

http://www.hydrogenaudio.org/forums/index....showtopic=16295

[lvqcl]

castanets2 sample encoded with falabaac and faac @ 128 kbps (encoding options: only -b 128)

[attachment=7277:castanet...falabaac.aac][attachment=7276:castanets2_faac.aac]

This post has been edited by lvqcl: Jan 3 2013, 19:41

[LithosZA]

castanets2 sample encoded with falabaac and faac:

Attached File castanets2_falabaac.aac ( 111.53K ) Number of downloads: 0
Attached File castanets2_faac.aac ( 111.28K ) Number of downloads: 0

Ok, falabaac lost with that sample.. I wonder if decreasing the bandwidth to the same bandwidth that FAAC uses would help?
Edit: Where can I find the castanets2 sample?

This post has been edited by LithosZA: Jan 3 2013, 19:52

[lvqcl]

Currently it can be downloaded here:

http://web.archive.org/web/20030618115117/...s/test_samples/

[LithosZA]

Currently it can be downloaded here:

http://web.archive.org/web/20030618115117/...s/test_samples/

Thanks, I quickly tried it out with 16Khz bandwidth and it didn't make a difference.

[blueheart73]

castanets2 sample encoded with falabaac and faac:

Attached File castanets2_falabaac.aac ( 111.53K ) Number of downloads: 0
Attached File castanets2_faac.aac ( 111.28K ) Number of downloads: 0

Ok, falabaac lost with that sample.. I wonder if decreasing the bandwidth to the same bandwidth that FAAC uses would help?
Edit: Where can I find the castanets2 sample?

yes, I tried this sample before and found that there will be pre-echo affect because using LONG_WINDOW. if you use option "-w 16" or use high bitrate to encode, the high frequency will be filterd, and maybe sound better.
And , to solve this quilkly change audio, you can use option "-t 1" (means time resolution first, I will use ONLY_SHORT_WINDOW to process this sample), and sound good!

But the seriously problem is that I didn't found the good method to switch the LONG_WINDOW and SHORT_WINDOW!!! the method in the ISO I tried , but it does;t work well!!
Can somebody teach me a good method(If can send a paper detail about it I will apprecaite for you)

[blueheart73]

One more I want to say is that, use "-t 1" option maybe have some problem or bugs in the low bitrate, the quantizer is not good, so this option is not recommed except for the sample only quickly change
And I tried FAAC in 96kbps, the FAAC limited the bandwidth to 10kHz. Besides, I analyzed the FAAC code and found that it always use SHORT_WINDOW to process the audio(no LONG_WINDOW, it canceled in some code),
So I am thinking if it faced the same problem with me ? I am not sure about that.

Can someone familiar with the AAC tell me how to resolve this problem and how to do quickly and stably window-switching method?

[jrmuizel]

Can someone familiar with the AAC tell me how to resolve this problem and how to do quickly and stably window-switching method?

Maybe look here:
https://github.com/mstorsjo/fdk-aac/blob/ma...lock_switch.cpp

or here:
Robust Block Switching Decision for Transform-based Audio Coder
http://citeseerx.ist.psu.edu/viewdoc/downl...p1&type=pdf

[blueheart73]

Can someone familiar with the AAC tell me how to resolve this problem and how to do quickly and stably window-switching method?

Maybe look here:
https://github.com/mstorsjo/fdk-aac/blob/ma...lock_switch.cpp

or here:
Robust Block Switching Decision for Transform-based Audio Coder
http://citeseerx.ist.psu.edu/viewdoc/downl...p1&type=pdf

thank you, I will try it in the next version and see the results!