Winamp FhG AAC encoder with gapless fix, Updates from the 5.62 release Options

[Gainless]

Foobar's ABX logs don't provide the information about DSP (Replaygain, etc.).
Your sample is ok. FhG LC-AAC doesn't put additional frames (delay) and volume is practically the same.

Fine, would have kinda sucked if I had to do it again

@ C.R. Heinrich

Will the encoder still be updated in the future, maybe with some more modes, or is this the final version?

[C.R.Helmrich]

Will the encoder still be updated in the future, maybe with some more modes, or is this the final version?

Quality-wise, we are not planning any huge updates, at least not for stereo and mono. But if there is a significant interest in other VBR modes, we will consider adding those.

P.S.: Helmrich, not Heinrich And you can call me Chris.

B66pak: Please keep in mind that MP3 is not AAC. And Fraunhofer's encoder is already faster than nero's, as Destroid posted. So I don't see the need to invest a lot of time in tuning our encoder for speed. In fact, we already did, so we won't be able to improve it by more than a few percent, I'd guess.


This post has been edited by C.R.Helmrich: Nov 14 2011, 14:48

[Destroid]

i can see that another fraunhofer tool (the mp3 encoder) is king of the speed...

Ah, the Fraunhofer MP3 Surround encoder. Actually that test in b66pak's link was CBR 128kbps. But here is its VBR results with the relevant contenders/affiliates:

CODE

fhg aac 3.2.3      --vbr 4   115s   10.10%
fhg aac 3.2.3      --vbr 5   125s   15.60%
helix mp3 5.1      -V120      29s   13.68%
fhg mp3s enc 1.5   -m 2       34s   13.65%

Comments:
- the "Process time" of the Fhg MP3 surround encoder was the same after ICCpatch-ing the executable, although the global times differed (obviously the unpatched time was slower)
- on this system the Helix encoder prevailed (for those wanting to know, material tested was hard rock CD from 1997 with total running time of 46m43s)

@C.R.Heineken - I wanted to ask inquire into the possibility of a preset between --vbr 4 and --vbr 5. I think --vbr 4 is a good preset for about average 130kbps, but the jump to > 200kbps with --vbr 5 seems a bit overkill, but I don't have any reason to doubt it's well tuned.

[Gainless]

Btw, how much listenings do I need in an ABX test to make a proper judgement about the encoded material? I ever do around 10-15 ones, but that gives me weird results, e.g. 9/11 on the one and 2/11 on the other hand...

PS:
Sorry to Chris for the spelling mistake^^

This post has been edited by Gainless: Nov 15 2011, 17:16

[C.R.Helmrich]

@C.R.Heineken - I wanted to ask inquire into the possibility of a preset between --vbr 4 and --vbr 5. I think --vbr 4 is a good preset for about average 130kbps, but the jump to > 200kbps with --vbr 5 seems a bit overkill.

True. Yes, as said, it's a possibility. I'll keep it in mind. Btw, do you - or anyone else - have speed measurements for the iTunes encoder (via qaac or qtaacenc maybe)?

Chris

[no404error]

Btw, do you - or anyone else - have speed measurements for the iTunes encoder (via qaac or qtaacenc maybe)?

Chris

Intel Pentium G630 / 2x4Gb DDRIII-1333 / ImDisk 1.5.3 / Windows 7 Home Premium x64

CODE

neroAacEnc 1.5.4.0

-q 0.00 - 92,7x
-q 0.25 - 43,5x
-q 0.50 - 39,1x
-q 0.75 - 38,1x
-q 1.00 - 37,2x

qaac 1.04, CoreAudioToolbox 7.9.7.3

--tvbr 0   -q 2 - 45.9x
--tvbr 36  -q 2 - 40.3x
--tvbr 63  -q 2 - 36.6x
--tvbr 100 -q 2 - 35.0x
--tvbr 127 -q 2 - 33.9x

fhgaacenc 20111104, enc_fhgaac 1.02

--vbr 1 - 67,9x
--vbr 2 - 42,7x
--vbr 3 - 53,3x
--vbr 4 - 52,5x
--vbr 5 - 48,6x

This post has been edited by no404error: Nov 16 2011, 06:19

[IgorC]

no404error
Nice results.

Just some observations.
TVBR has no HE-AAC mode. It will be more correct to compare HE-AAC vs HE-AAC or LC-AAC vs LC-AAC.
Apple HE-AAC (--cvbr 64 --he) vs FhG HE-AAC (--vbr 2). Also qaac -q2 and -q1 has practicly identical quality. The difference isn't perceptible.

[C.R.Helmrich]

Thanks a lot, 404!

Also qaac -q2 and -q1 has practicly identical quality. The difference isn't perceptible.

Which setting was used in the last listening test? qtaacenc seems to default to -high, is that the same as -q 2 in qaac? And does nero automatically switch to HE-AAC for low -q values?

Chris

[IgorC]

Which setting was used in the last listening test?

qtaacenc --highest (equivalent to qaac -q 2)

[no404error]

no404error
Nice results.

Just some observations.
TVBR has no HE-AAC mode.

I now. But I test 5 to 5 default profiles, Apple Core Audio AAC vs Fraunhofer AAC. FhG - winner.

TVBR has no HE-AAC mode. It will be more correct to compare HE-AAC vs HE-AAC or LC-AAC vs LC-AAC.

CODE

qaac 1.04, CoreAudioToolbox 7.9.7.3

Preset
Min    (--tvbr   0) - -q 1/2 - 58.3x/45.9x
Low    (--tvbr  32) - -q 1/2 - 54.2x/40.3x
Medium (--tvbr  64) - -q 1/2 - 48.9x/36.6x
High   (--tvbr  96) - -q 1/2 - 46.5x/35.0x
Max    (--tvbr 127) - -q 1/2 - 43.8x/33.9x
   (--cvbr 64 --he) - -q 1/2 - 42.2x/36.1x

fhgaacenc 20111104, enc_fhgaac 1.02

Preset
--vbr 1 - 67,9x
--vbr 2 - 42,7x
--vbr 3 - 53,3x
--vbr 4 - 52,5x
--vbr 5 - 48,6x

Core Audio AAC HE mode is a spherical cow IMHO

Also qaac -q2 and -q1 has practicly identical quality. The difference isn't perceptible.

As I see, q1 has unacceptable higher bitrate than q2. Up to 5%

And does nero automatically switch to HE-AAC for low -q values?

-q 0.00-0.15 - HEv2
-q 0.16-0.30 - HEv1
-q 0.31-1.00 - LC


This post has been edited by no404error: Nov 16 2011, 22:56

[Destroid]

Btw, do you - or anyone else - have speed measurements for the iTunes encoder (via qaac or qtaacenc maybe)?

Finally got the hang of QAAC+[Apple Application Support] but doesn't reveal much else already posted (thanks no404error for the fast results), so figured I'd at least throw in FAAC results too:

CODE

fhg aac 3.2.3       --vbr 4   115s   10.10%
fhg aac 3.2.3       --vbr 5   125s   15.60%
neroaac 1.0.7.0       -q0.5   190s   12.18%
neroaac 1.5.4.0       -q0.5   146s   12.63%
qaac 1.04*       --cvbr 170   158s   11.78%
qaac 1.04*        --tvbr 82   159s   11.58%
qaac 1.04*        --tvbr 91   160s   14.12%
faac 1.28 mod -b170 -c19600   174s   12.04%
faac 1.28 mod         -q130   172s   11.67%

*QT 7.7.1 CoreAudioToolbox 7.9.7.8

Comments:
- all tests were on same machine with same material, however prior tested codecs' results are copy+pasted to this table
- Apple TVBR also has gaps in the scale (i.e. no stop points between --tvbr 82 and --tvbr 91)
- FAAC ABR 170 defaulted to 20000Hz bandwidth so I lowered it to 19600 (same bandwidth as FAAC -q130)
- at first glance FAAC ABR appeared not to "flex" much but closer examination revealed bitrate rapidly fluctuates 150-210kbps

[IgorC]

But I test 5 to 5 default profiles

The thing is qaac's default is q2 and iTunes q1.

Apple Core Audio AAC vs Fraunhofer AAC. FhG - winner.

Winner for what? For speed or speed/quality/bitrate?
Sorry, I don't see how You've managed to make any conclusion.
I can add almost useless -q0 to LAME and claim that's it is slowest MP3 encoder and Helix is a winner (?).

Let's define the relations ( with acceptable level of precision): speed/quality/bitrate and then talk about winners.

What the goal of comparing fhg -v 3 and --tvbr 64 or 96 if these settings haven't even the same bitrate? Variation of all three variables gives us ....nothing.

Who does guarantee You that FhG is any better in quality terms or at least on par with tvbr -q0 at the same bitrate?

This post has been edited by IgorC: Nov 17 2011, 01:08

[no404error]

Winner for what? For speed or speed/quality/bitrate?

For speed. Only speed test at this time.

Who does guarantee You that FhG is any better in quality terms or at least on par with tvbr -q0 at the same bitrate?

My ears and... Your own Public AAC Listening Test Only samples with voice is important for me. FhG faster, cleaner and much hardware compatibility. Winner, at least only for me

[C.R.Helmrich]

FhG faster, cleaner and much hardware compatibility.

The hardware compatibility of all LC-AAC/HE-AAC encoders should be the same, since all implement the exact same specification. So the Fraunhofer encoder has no advantage here.

By the way, the Fraunhofer encoder in Winamp runs in "high quality" mode, which in case of VBR means the highest quality available. In "normal quality" mode, the encoder sounds slightly (maybe indistinguishably?) worse but is significantly faster. So comparing against QuickTime -q 2 is appropriate.

Anyway: thanks Destroid and (again) no404error for the analyses!

Chris

This post has been edited by C.R.Helmrich: Nov 17 2011, 21:37

[Gainless]

FhG faster, cleaner and much hardware compatibility.

The hardware compatibility of all LC-AAC/HE-AAC encoders should be the same, since all implement the exact same specification. So the Fraunhofer encoder has no advantage here.

By the way, the Fraunhofer encoder in Winamp runs in "high quality" mode, which in case of VBR means the highest quality available. In "normal quality" mode, the encoder sounds slightly (maybe indistinguishably?) worse but is significantly faster. So comparing against QuickTime -q 2 is appropriate.

Anyway: thanks Destroid and (again) no404error for the analyses!

Chris

Can you give us the command line for that quality switch?

[C.R.Helmrich]

Can you give us the command line for that quality switch?

Sorry, no. I didn't write enc_fhgaac.dll and the command-line wrapper. And to be clear: the command-line encoder you are talking about is neither developed nor supported nor endorsed by Fraunhofer (Fraunhofer sells its own command-line encoder which has dozens of switches, incl. quality). Nullsoft, in particular benski, have put the encoder and a lot of work into Winamp so that people like and use their software. So please appreciate that by using the encoder through Winamp.

Chris

[Destroid]

By the way, the Fraunhofer encoder in Winamp runs in "high quality" mode, which in case of VBR means the highest quality available. In "normal quality" mode, the encoder sounds slightly (maybe indistinguishably?) worse but is significantly faster. So comparing against QuickTime -q 2 is appropriate.

I have to admit that I don't see any quality settings outside the VBR slider (kbps) Maybe it's totally hidden :shrug:

As for my QAAC results, the -q 2 setting was defaulted (quality 96) and of course lower -q settings were faster:
-q 2 (96) = 17.4x
-q 1 (64) = 24.7x

edit: doh! should have timed Fhg encoder via Winamp before posting
Test result: --vbr 4 = same speed as commandline

This post has been edited by Destroid: Nov 18 2011, 01:20

[IgorC]

By the way, the Fraunhofer encoder in Winamp runs in "high quality" mode, which in case of VBR means the highest quality available. In "normal quality" mode, the encoder sounds slightly (maybe indistinguishably?) worse but is significantly faster. So comparing against QuickTime -q 2 is appropriate.

Chris,

There is no absolutely any information how changes the quality of Apple and FhG encoders with different speed settings.
Apple and FhG are totally different encoders and there is no parallel of Apple's -q2/q1 and FhG -high/normal quality.

There is already unfortunate myth in community of video compression that Xvid (MPEG4 ASP) is faster than x264. And it's simply not true. x264 with fast presets is still faster and better than XviD. Let's not create other myth.

P.S. Is speed still has any priority when CPUs are very fast and getting even more faster nowadays? At least it's not critical anymore.
Last time I've checked people preferred to squeeze the last few bits at cost of slower speed. http://www.hydrogenaudio.org/forums/index....c=58731&hl=

So please appreciate that by using the encoder through Winamp.

I do appreciate a lot. I still can't get that FhG (at least its particular Winamp's edition) encoder is available publicly. Thank You, Guys.

This post has been edited by IgorC: Nov 18 2011, 03:55

[C.R.Helmrich]

I have to admit that I don't see any quality settings outside the VBR slider (kbps) Maybe it's totally hidden :shrug:

P.S. Is speed still has any priority when CPUs are very fast and getting even more faster nowadays? At least it's not critical anymore.

Indeed. I think there's no speed slider in Winamp's AAC encoder because there's no need for it. There's no point in encoding faster=worse if at high quality you're still at least 40x faster than real-time.

Chris

[lvqcl]

On the other hand, Winamp Format Converter is still single-threaded. And if a user have 4-core CPU then Winamp encoder is quite slow compared with e.g. foobar2000 + 4 NeroAacEnc encoders in parallel.

[IgorC]

There is a new version of Winamp with updated FhG AAC encoder 3.2.4

Last versions of FhG HE-AAC produce comparable bitrate at ~64 kbps as Apple HE-AAC encoder. So I've tried these encoder for set of samples. Won't go into details just will mention that FhG was quite better.

Good job indeed.

This makes the last public test at 64 kbps obsolete though (in less than one year).

[b66pak]

There is a new version of Winamp with updated FhG AAC encoder 3.2.4

where?
_

[lvqcl]

http://forums.winamp.com/showthread.php?t=332010&page=3

Winamp 5.623 Released

First post has been updated accordingly.

Note that this is a forum exclusive (main site update on Monday 12th Dec).
Though all builds in all flavors & languages are up on the server, if you know the filenames...

Early Xmas present :-)

[b66pak]

thanks a lot...
_

[C.R.Helmrich]

Last versions of FhG HE-AAC produce comparable bitrate at ~64 kbps as Apple HE-AAC encoder. So I've tried these encoder for set of samples. Won't go into details just will mention that FhG was quite better.

Good job indeed.

Thanks By "last versions", do you mean 3.2.3 and later? Were the average bit-rates of version 3.2.2 (the one of the 96-kb test) higher or lower than those of the Apple encoder on your test set at ~64 kb?

Chris

This post has been edited by C.R.Helmrich: Dec 11 2011, 22:48