Skip to main content
Topic: AAC FDK/FHG, is it really the Future of AAC? (Read 12035 times) previous topic - next topic
0 Members and 2 Guests are viewing this topic.

AAC FDK/FHG, is it really the Future of AAC?

After I heard about Fraunhofer's codec interations (Fhg/Fdk) Being a next step/evolution in quality in the Realm of AAC, which is my favorite lossy codec, I thought it was about time to switch from Nero which was pretty much dead.

I've always liked Nero a lot and wholeheartedly adopted it about 6 years ago to encode all my portable and even part of my permanent collection in it after it surpassed aoTuV (my previous lossy choice after LAME) in both quality and compatibility.

Back at the day, I did several ABX tests to find my personal sweet spot for both non-annoying-low-size-portable-listening-quality (back in the day 8gb Micro SD cards were still expensive) and personal transparent archival setting. Since then I've been using the same settings on nero:
Q0.4~0.45 for portable listening, which was near transparent to me, some easy samples I would even fail on ABX at 0.45...
Q0.5~0.6 for storage/computer-quiet listening, which gave between 170~185kbps. Besides a single killer-sample, I always failed abx at this preset with Nero.

So after I heard FHG was considerable better and more optimized than NERO, I was hoping to at least encode it at similar to lower presets than this.
My first surprise was of course, to discover that Winamp FHG's doesn´t even Have a 160kbps preset. It jumps from 128 to 192, which I found absurd. Angry having installed winamp for nothing i switched to FDK.
I also got a little frustrated by the lack of a middle therm between 160 and 192 kbps settings, 190 (V2) setting is what I used with lame for 100% transparency, I also think compatibility is very important, if a format does not have any perceiveable gain in quality towards an old more compatible one, then I should just stick with the old one for it's compatibility.

Anyway, I decided to test both the 144 and 160 presets In Fdk with big hopes to fail ABX tests at even 144kbps which would result in significant savings in space, specially on mobile on the run.

I first tried a couple of "normal", not particularly difficult samples, encoded and before even started the ABX tests, on the first time listening to the freshly encoded file I could already hear easily distinguishable high-pitched artifacts at splashes/hit-hats and echos. I promptly set up the ABX test hoping it was just placebo and unfortunately I didn´t have much difficult in confirming it really was some real compressing artifacts I was hearing.

This is the first 144 log:

Code: [Select]
foo_abx 1.3.4 report
foobar2000 v1.3.2
2014/06/29 12:05:27

File A: F:\Musicas\ANIMES e Jmusic\Evangelion\Rebuild of Evangelion - EVANGELION 2.0 YOU CAN (NOT) ADVANCE\DISC II\15 The Final Decision We All Must Take (2EM33).flac
File B: M:\EVA AAC\Rebuild of Evangelion - EVANGELION 2.0 YOU CAN (NOT) ADVANCE\DISC II\15 The Final Decision We All Must Take (2EM33).m4a

12:05:27 : Test started.
12:14:42 : 01/01  50.0%
12:14:49 : 02/02  25.0%
12:14:59 : 03/03  12.5%
12:15:07 : 04/04  6.3%
12:15:13 : 05/05  3.1%
12:15:20 : 06/06  1.6%
12:15:32 : 07/07  0.8%
12:15:42 : 08/08  0.4%
12:15:49 : 09/09  0.2%
12:15:56 : 10/10  0.1%
12:15:59 : Test finished.

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


I then decided to try encoding at 160kbps, hoping I would fail an ABX at this level, this is the result:

Code: [Select]
foo_abx 1.3.4 report
foobar2000 v1.3.2
2014/06/29 12:31:42

File A: F:\Musicas\ANIMES e Jmusic\Evangelion\Rebuild of Evangelion - EVANGELION 2.0 YOU CAN (NOT) ADVANCE\DISC II\15 The Final Decision We All Must Take (2EM33).flac
File B: C:\Users\XXXX\Desktop\15 The Final Decision We All Must Take (2EM33).m4a

12:31:42 : Test started.
12:33:25 : 01/01  50.0%
12:33:30 : 02/02  25.0%
12:33:36 : 03/03  12.5%
12:33:40 : 04/04  6.3%
12:33:59 : 05/05  3.1%
12:34:09 : 06/06  1.6%
12:34:16 : 07/07  0.8%
12:34:28 : 08/08  0.4%
12:34:33 : 09/09  0.2%
12:34:37 : 10/10  0.1%
12:34:39 : Test finished.

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


Really disappointing.
I then tried Nero Q0.5 setting, the old setting I'd have used, it gave about 179kbps for this particular sample and I tried abxing it, had to focus really hard, thought I was onto some artifacts but ultimately failed the ABX.
I should of course try and lower the Nero setting to about the same bitrate of the FDK encode but i was already getting tired so I called out for the day. (I usually get tired pretty quickly when abxing, can hardly do it for more than 20 minutes on difficult samples)

Then today, I decided to try FDK again and I went on to grab a quieter and suposedely easier song and try testing at 160kbps again, while at a first casual listening I wasn´t easily identifying any artifacts, so it wasn´t annoying, at further ABX inspection with a little bit of focus I was able to pinpoint more than one excerpts of the song which I could differentiate from the lossless version:

Code: [Select]
foo_abx 1.3.4 report
foobar2000 v1.3.2
2014/06/30 12:43:32

File A: F:\Musicas\ANIMES e Jmusic\Evangelion\Rebuild of Evangelion - EVANGELION 2.0 YOU CAN (NOT) ADVANCE\DISC II\21 Tsubasa wo Kudasai -Tribute to 'Sound of Music'.flac
File B: C:\Users\XXX\Desktop\21 Tsubasa wo Kudasai -Tribute to 'Sound of Music'.m4a

12:43:32 : Test started.
12:43:44 : 01/01  50.0%
12:43:48 : 02/02  25.0%
12:44:00 : 03/03  12.5%
12:44:14 : 04/04  6.3%
12:44:24 : 05/05  3.1%
12:44:33 : 05/06  10.9%
12:44:40 : 06/07  6.3%
12:44:49 : 07/08  3.5%
12:44:58 : 08/09  2.0%
12:45:08 : 09/10  1.1%
12:45:17 : 10/11  0.6%
12:45:19 : Test finished.

----------
Total: 10/11 (0.6%)


So, it fails my transparency target but I admit it was a very difficult test and the file wasn´t annoying at all.
But I'm not a purist when It comes to audio encoding and I usually prefer to store most of my music, specially if they are not my personally favorites or exceptional composition in Lossy format anyway, so I would like to at least find the transparency threshold for a given codec to do it. Now I don´t feel comfortable at all on encoding to lossy my music at 160 kbps, and without a intermediate preset bellow 192kbps, I feel like I'll be actually downgrading quality x size wise from switching from nero which I felt comfortable at around 180kbps.

Another problem with FDK that kinda of got to me, is the fact that even if I choose to stick up with 192 kbps encodes, that meant having a harder more conservative lowpass setting. Nero used 18.5 khz lowpass, while I found out that FDK at 192 uses 17khz and 16 at 160 which honestly I find it rather low. At 224kbps it uses a 20khz lowpass witch is reasonably so why such an steep move from 20 to 17khz at just one lower VBR step? Others codecs have a more subtle Lowpass vs VBR-setting curve. aoTuV had a much more optimized high level lowpass 20khz at even lower bitrates, and even LAME have a 18.5khz lowpass at V2 (although if I recall correctly, it is a "hybrid" low pass, were it will only give extra bitrate if it has enough to spare, right?)
Of course I understand that above 16khz most of the audio is actually noise and even tough I can hear sine-waves at 19khz, It doesn´t mean I would be able to hear low-volume overtones at this level, let alone dream on abxing it. Yet I still feel uncomfortable of deleting my FLACs and embracing FDK as a lossy-archival solution knowing that at 160kbps my files would only top at 16khz, that is almost like encoding a wave at 32khz sample rate! It just feels wrong!

Does anyone knows about other tests between NERO and FDK ever been conducted? I have mentioned I use nero in other places recently just to be bashed about how "Inferior and abandoned" it is today, so when I heard all the hype towards FHG I got quite disappointed. I am still considering on doing more tests with more samples, (specifically, a direct nero vs fdk at the same final bitrate abx.) but since my first impression is already bad I decided to post here and listen to you guys opinions first. Does anyone know if FDK will still receive tuning in the future with an intermediate setting above 160kbps and why the conservative/steep low-pass settings?
Another question: I've always heard that we should avoid CBR settings as they are vastly inferior than an equivalent VBR setting, but in this case wouldn´t it be interesting to try a 175kbps CBR setting for the Winamp FHG codec or will it definetely be inferior than the q0.x settings of Nero at the same size?

Please excuse if I did anything wrong, I understand that more testings should be done, but I'd like to see other people tests first, specially if they are extensive. Also I wasn´t really sure if this was supposed to go on the AAC-tech or the listening tests section. I usually only lurk on HA.
Cheers

AAC FDK/FHG, is it really the Future of AAC?

Reply #1
My opinion, why would they make FDK (open source) "good" "their best" when they still sell FHG (licensed)?

AAC FDK/FHG, is it really the Future of AAC?

Reply #2
FHG (as in Winamp) *is* their licensed code isn't it?


AAC FDK/FHG, is it really the Future of AAC?

Reply #4
Interesting.  Did you use FhG encoder from Winamp, or FDK encoder?  You mention both but apparently they aren't the same:  http://www.hydrogenaud.io/forums/index.php?showtopic=95989.  Maybe you should try the Apple AAC encoder via qaac.

AAC FDK/FHG, is it really the Future of AAC?

Reply #5
Interesting.  Did you use FhG encoder from Winamp, or FDK encoder?  You mention both but apparently they aren't the same:  http://www.hydrogenaud.io/forums/index.php?showtopic=95989.  Maybe you should try the Apple AAC encoder via qaac.


Yes, I already know they aren´t the same, as a matter of fact if you read carefully my post again you will see I mention I was first going to test Fhg (winamp´s encoder) and once discovering It did´t have a  intermediate preset between 128 and 192 VBR settings I got angry with it and decided to just test FDK.

I read that particular thread you posted before, but there aren´t many thorough information and tests between the differences between FHG and FDK yet. After I posted my OP, I decided to further investigate the Low-pass issue that I described from FDK.
Here are some interesting findings:



FDK, FHG and Nero side by side.

As you can see, FHG lowpass implementation is very different to FDK which seems like a simple crude and hard low-pass. FHG seems like a more intelligent algorithm, Starting at 16khz by each 500hz it furthers lowers it sensibility to preserve bitrate and don´t completely cut off high signals above 19khz. FDK lowpass is much more aggressive. FHG looks similar to Nero's Lowpass.
So yes, not only it's different on the VBR settings, FHG clearly has further optimizations on it's encoder.
(Yes, I do realize this picture don´t really prove FDK is inferior, it could means the opposite, more bitrate at lower frequencies and thus actually achieving better results. But by now we all know this is hardly the case, as eahm cleverly stated: There is hardly any point in Fraunhofer making the free FDK iteration superior to it's commercial FHG.)
Sadly however, I've almost always found Q0.55(this picture) on NERO transparent, so an ABX test wouldn´t be much usefull.  it would be interesting to try and do some ABX with fhg and fdk at 192kbps, but I guess both will sound transparent anyway. (Tough by using DSPs to increase the stereo separation, I've often found thins an interesting way to further extend the limit of transparency of high frequency transcoded files, yet this is not really scientific as it hurts the psycho acoustic model) Nevertheless If I were to choose one of these 3 formats to use at 192kbps, I'd either go with FHG or Nero, probably nero at a lower q, anyway.
I'm considering on doing some personal abx's at 128kbps with both FHG and Nero to see if I can spot which one is better. 128 abx was always easy to me with nero, I have a feeling it won´t be more difficult with fhg.

By the way, I realized that the people that actually criticized Nero for not being updated for a long time and being "abandoned" ironically forgot that LAME hasn´t really being fine-tuned for quality for over 4 years now too. Yet it still is used and competitive at higher bitrates. I guess the hype of FHG as well as opus is due to it's superior performance at lower bitrates (less than 128), but at higher bitrates, they haven´t really significantly surpassed NERO. At least no test ever corroborated that.
There has been some time since HA had a public's test, but other sites (although less reliable) have put Nero head to head with FHG even as this very day.

EDIT:
Quote
Maybe you should try the Apple AAC encoder via qaac.

I don´t really like apple's products and their patent savvy politics toward closed software. I Heard it's possible to use QAAC without installing the bloated itunes, but is it possible without installing quicktime? For testing sake I guess I could consider installing QT just for performing some quick tests. But I believe in the end the difference would me marginal and I'd stick with either nero or winamp for conveniently staying away from apple.

AAC FDK/FHG, is it really the Future of AAC?

Reply #6
Yes, I missed where you switched from FhG to FDK.

Anyway, if you want to try the Apple encoder, download 7zip, the latest iTunes installer and qaac and makeportable.zip from https://sites.google.com/site/qaacpage/cabinet.  No need to install iTunes; makeportable will extract the necessary files from iTunes which you can then delete.


AAC FDK/FHG, is it really the Future of AAC?

Reply #8
A listening test comparing FDK AAC against FhG, Nero and maybe LAME would be nice. Just so that we know where FDK really fits in. I am pretty sure FDK outperforms FAAC.

AAC FDK/FHG, is it really the Future of AAC?

Reply #9
Not including FDK (or FAAC although you can certainly find tests including that), but a fairly recent listening test at 96kbps is one of the reasons that Nero is considered obsolete.  It came last and there is hardly any way it would have improved since.  CT is available as aacplusenc.

AAC FDK/FHG, is it really the Future of AAC?

Reply #10
Yes, I missed where you switched from FhG to FDK.

Anyway, if you want to try the Apple encoder, download 7zip, the latest iTunes installer and qaac and makeportable.zip from https://sites.google.com/site/qaacpage/cabinet.  No need to install iTunes; makeportable will extract the necessary files from iTunes which you can then delete.


I guess I'll just select some samples and borrow my father's notebook which is an Imac when I visit him this weekend and encode the files there for the listening test.

Not including FDK (or FAAC although you can certainly find tests including that), but a fairly recent listening test at 96kbps is one of the reasons that Nero is considered obsolete.  It came last and there is hardly any way it would have improved since.  CT is available as aacplusenc.


Wasn´t nero thoroughly optimized to it's default q0.5 setting? Performing poorly at 96 kbps doesn´t really imply it'll also be equally inferior at higher bitrates, specially on it's most optimized interval: q0.4 to q0.6. AFAIK, different optimizations, even on the same lossy technology, can lead to different levels of quality. For using at lower bitrate, of course I wouldn´t even consider using Nero. When I first looked into FhG some weeks ago I saw that test and didn´t get much impressed since I consider 96kbps pretty low for my own standards to anything that isn´t Video Soundtrack. I tried looking for 128kbits tests but aside from SoundExpert.org public tests which rate Itunes AAC+ at first and Nero second, (lolwut?) I couldn´t find any more reliable tests to base On. That Is why I opened this thread in the first place.

I did perform some of my own comparisons at 48 and 64kbps of HE-AAC from FDK (since it was added for Handbrake) and Opus for video encoding purposes and got pretty impressed with Opus and Fdk compared to Nero (it was clearly superior to nero), but I'd never use HE for music, not even on portable devices. And for that matter, I consider 96 and 112kbps pretty low too for most samples and types of music. (Of course there are exceptions).

A listening test comparing FDK AAC against FhG, Nero and maybe LAME would be nice. Just so that we know where FDK really fits in. I am pretty sure FDK outperforms FAAC.



Yes, I'm also thinking is worthy to try and give a Shot. While I'm not experienced as other veterans of the forum, I think is not a bad idead to try it myself, so I decided to organize my own significant test and comparisons of AAC codecs after I created this thread.
I'll decide between 4 to 5 different samples and test then with FHG, Nero, FDK and maybe Itunes at 128 and 160 kbps. (not much point on testing 192kbps since I would probably fail all tests). I'll report to you guys as long as I perform the tests. (I'll probably have to split the ABX between a couple of days and hours because I get tired pretty easily)
Do you guys have any suggestion regarding samples I could use? I think One particular Killer sample should be interesting. I was think about one of the killer samples uploaded by /mnt
Also, Is any it worthy to use FhG at CBR since it doesn´t have a 160 vbr setting? I've read that Itunes and Fhg CVBR are pretty reliable, what you guys think?


AAC FDK/FHG, is it really the Future of AAC?

Reply #11
You've probably seen this, but it is an FDK @ 128 test:
http://www.hydrogenaud.io/forums/index.php?showtopic=104471

When you start to look for very precise listening test parameters, really the only thing that matters is what you hear.

Or see here for a 128 kbps test including both Winamp and libfdk AAC
http://soundexpert.org/encoders-128-kbps

AAC FDK/FHG, is it really the Future of AAC?

Reply #12
You've probably seen this, but it is an FDK @ 128 test:
http://www.hydrogenaud.io/forums/index.php?showtopic=104471

When you start to look for very precise listening test parameters, really the only thing that matters is what you hear.

Or see here for a 128 kbps test including both Winamp and libfdk AAC
http://soundexpert.org/encoders-128-kbps

Dude, as IgorC already told you, stop quoting SoundExpert's tests.

In the first test you linked the other AAC is FAAC not any of the newer ones.

AAC FDK/FHG, is it really the Future of AAC?

Reply #13
Now I'm getting slated for posting stuff before I even knew not to?  I'm not psychic!  I know what was in the test.  I never said it was perfect, but a test that include FDK-AAC.  Anyway, Kamedo hasn't tested the encoders that sephirotic is interested in, so I did my best.  If you want to get a hard-on about that, be my guest.

AAC FDK/FHG, is it really the Future of AAC?

Reply #14
It's not that they are not perfect, they are not valid.

AAC FDK/FHG, is it really the Future of AAC?

Reply #15
So, I'm in the process of selecting the samples for this small test of mine. I already choose 2 difficult samples from this forum musorgsky_bydlo, and KMFDM_DOGMA and decided to add some of my own since this is a personal test with musics I like.

First Sample I selected and Uploaded it's here.
it's the final 30s of Mahler's 8th symphony. I think this turned out as a very interesting and useful difficult sample. Lot's of Brass, high pitched violins, both male and female voice with an organ in the background! Only missing for this being a perfect romantic-orchestral sample is a symphonic cymbal, too bad!

I think I'll also add The girlsdemo Crow Song from marina, which proved to be a difficult sample in previous tests I did recently (and the third genre of music I like the most after classical and jazz is J-pop, Yeah, don´t say it.)

While first trimming, transcoding and early listening for the test files I could already easily perceive that unfortunately Nero was suffering at Q0.4 to maintain a good quality. Particularly at 12s as we can hear a very high note from the Soprano. A quick ABX test and I could even distinguish a direct Lossy to lossy comparison with CVBR Fhg @ 128!. However, Fhg wasn´t really much better as I would'nt give more than 3 rating to it either.
I'll post the full ABX logs once I select all the samples and start the proper ABX tests, this was just a quick early listening for this particular sample.

Now I'm getting slated for posting stuff before I even knew not to?  I'm not psychic!  I know what was in the test.  I never said it was perfect, but a test that include FDK-AAC.  Anyway, Kamedo hasn't tested the encoders that sephirotic is interested in, so I did my best.  If you want to get a hard-on about that, be my guest.

Well, I don´t know all the story between SoundExpert methodology and the Hydrogen Audio stand against them, but when I first saw that 128 test with an AAC+ with 4+ minimal rating topping the chart It already smelled pretty fishy to me. And I also don´t get their "20 point scale" since the base score you can give to samples is 1 to 5. Perhaps if eahm could be courteous enough to elaborate for us?
The Kamedo2's test you posted comparing FDK with FAAC and Mp3 is very interesting, too bad Fhg or nero wasn´t tested too!

AAC FDK/FHG, is it really the Future of AAC?

Reply #16
Perhaps if eahm could be courteous enough to elaborate for us?

I won't, don't care, I have warnings and shit because I keep talking bad about it etc. so... no.

I gathered sample files/killer sample files for ABX tests, you may find this pack useful: https://db.tt/BAwUquDG

AAC FDK/FHG, is it really the Future of AAC?

Reply #17
So, I started testing 3 samples I selected, first to try and find the Lower limit of what I consider transparent for both Nero and Fhg using CVBR.
Again I confirmed for Nero that at Q0.5 I failed at pretty much every sample. With Fhg, the threshold is slightly lower. While I did manage to ABX at around 160kbps CVBR some samples, others I failed, ultimately at 176kbps CVBR (around the same bitrate for NERO), I failed all the 5 samples I tried to ABX with FHG, with the exception of one:

I still don´t have sorted logs for the variety of preliminary tests I performed with the exception of logs for the famous killer sample: "Show_Me_Your_Spine_Sample_(pre-echo).flac" submited by /mnt

With "Show Me your spine", I'm able to Even ABX nero at up to q0.6 albeit with lots of effort. I'd probably fail this test if it was an ABC/HR. Nevertheless is interesting to see that transient artifacts are generated at such high bitrates with nero too.
Testing with the FDK I was also able to ABX 'show me your spine' at the 192kbps setting. I found it was even more difficult than Nero and had to stop the test since I was getting tired, after a while I redid the test but I can´t say for sure it's really better than nero.

Logs:
Nero
Code: [Select]
foo_abx 1.3.4 report
foobar2000 v1.3.2
2014/07/02 15:51:11

File A: C:\Users\XXX\Desktop\sample selection\kiler samples (most pre-echo lame)\Show_Me_Your_Spine__Sample_(pre-echo).flac
File B: C:\Users\XXX\Desktop\sample selection\kiler samples (most pre-echo lame)\Show_Me_Your_Spine__Sample_(pre-echo)nero q0.6.m4a

15:51:11 : Test started.
15:52:10 : 01/01  50.0%
15:52:23 : 02/02  25.0%
15:52:28 : 03/03  12.5%
15:52:36 : 04/04  6.3%
15:52:51 : 05/05  3.1%
15:52:57 : 05/06  10.9%
15:53:07 : 05/07  22.7%
15:53:16 : 06/08  14.5%
15:53:22 : 07/09  9.0%
15:53:44 : 08/10  5.5%
15:53:52 : 09/11  3.3%
15:53:59 : 10/12  1.9%
15:54:18 : 11/13  1.1%
15:54:35 : 12/14  0.6%
15:54:38 : Test finished.

----------
Total: 12/14 (0.6%)


FDK
Code: [Select]
foo_abx 1.3.4 report
foobar2000 v1.3.2
2014/07/02 15:58:12

File A: C:\Users\XXX\Desktop\sample selection\kiler samples (most pre-echo lame)\Show_Me_Your_Spine__Sample_(pre-echo).flac
File B: C:\Users\XXX\Desktop\sample selection\kiler samples (most pre-echo lame)\Show_Me_Your_Spine__Sample_(pre-echo)fdk192.m4a

15:58:12 : Test started.
15:58:40 : 01/01  50.0%
15:59:20 : 02/02  25.0%
15:59:27 : 03/03  12.5%
15:59:34 : 04/04  6.3%
15:59:45 : 05/05  3.1%
15:59:53 : 06/06  1.6%
15:59:58 : 07/07  0.8%
16:00:02 : 08/08  0.4%
16:00:07 : 09/09  0.2%
16:00:21 : 10/10  0.1%
16:00:22 : Test finished.

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


And finally the FHG test @ 192kbps VBR

At first I was having some difficult with the "bat-like clicks" I was accustomed to hear with other codecs with this sample, Then while choosing an interval for the ABX test I found a very nasty very distinctive high-frequency artifact. VERY CLEAR.

Here is the log for the ABX:
Code: [Select]
foo_abx 1.3.4 report
foobar2000 v1.3.2
2014/07/02 14:43:51

File A: C:\Users\XXX\Desktop\sample selection\kiler samples (most pre-echo lame)\Show_Me_Your_Spine__Sample_(pre-echo).flac
File B: C:\Users\XXX\Desktop\sample selection\kiler samples (most pre-echo lame)\Show_Me_Your_Spine__Sample_(pre-echo)fhg192.m4a

14:43:51 : Test started.
14:44:28 : 01/01  50.0%
14:44:31 : 02/02  25.0%
14:44:35 : 03/03  12.5%
14:44:40 : 04/04  6.3%
14:44:42 : 05/05  3.1%
14:44:45 : 06/06  1.6%
14:44:47 : 07/07  0.8%
14:44:52 : 08/08  0.4%
14:44:56 : 09/09  0.2%
14:44:59 : 10/10  0.1%
14:45:02 : 11/11  0.0%
14:45:05 : 12/12  0.0%
14:45:07 : 13/13  0.0%
14:45:10 : 14/14  0.0%
14:45:12 : 15/15  0.0%
14:45:14 : 16/16  0.0%
14:45:17 : 17/17  0.0%
14:45:19 : 18/18  0.0%
14:45:22 : 19/19  0.0%
14:45:24 : 20/20  0.0%
14:45:27 : 21/21  0.0%
14:45:29 : Test finished.

----------
Total: 21/21 (0.0%)


I managed to even isolate the artifact on audacity and discovered it's peak at about 12khz frequency. For curiosity I took a look in the spectograph and this is how the artifact looks like:



This is a 100ms excerpt, Left Original, Middle FHG, Right Nero.
We can See two distinctive extended gains on 12 and 16khz frequency. Alto ugh I believe the one I was hearing was, of course, the lower 12khz.

A quick sleuthing and I was able to find another one on at least another point in a few seconds ahead of the test file. If anyone is interested and have difficulty replicating this finding, I can upload my encoded files.

Anyway, I was willing to admit FHG was superior to Nero at med-to-high bitrates due to its sounding better at CVBR 128 than nero at q0.4 on my early tests, however, at higher bitrates I haven´t been confident of it's superiority yet.
So far, the 160kbps tests have been very tiresome due to their increased difficulty. Out of the 4 samples I tested so far, 2 I was able to abx albeit with several difficulty between the original lossless with both nero, FDK and FHG, and the other two I've failed with both codecs. The 2 samples I was able to ABX was always the first samples being tested in a batch and the difference were so subtle It became pretty obvious I'd never be able to test them at an ABC/HR test, let alone establish what format is better at this same given bitrate. After this difficult beginning of testings, I decided that testing at 160kpbs level won´t provide any useful data so I'll stick only with the 128 kbps test. However, at 128kbps, I have already encountered samples were Fhg had notably less high pitch artifacts than Nero and sounding better. But then again, I never encode at 128.
I was willing to consider encoding with 192VBR Fhg that was when I stumbled on the previous described artifacts with FHG of the Killer sample "Show me your spine", I've reluctantly retracted that premise now. I think I'm going to stick with Nero and Q0.5 as I've done in the past and wait Fhg to mature. Or at least, till Winamp Is finally transitioned to it's new owner and starts to deliver updated versions of the Fhg codec.

I'll still perform a small 128kbps Listening test tough and post the results.

Perhaps if eahm could be courteous enough to elaborate for us?

I won't, don't care, I have warnings and shit because I keep talking bad about it etc. so... no.

I gathered sample files/killer sample files for ABX tests, you may find this pack useful: https://db.tt/BAwUquDG

Thanks for the samples, good material.

AAC FDK/FHG, is it really the Future of AAC?

Reply #18
I managed to even isolate the artifact on audacity and discovered it's peak at about 12khz frequency. For curiosity I took a look in the spectograph and this is how the artifact looks like:

FFT/window size looks too big for these pictures.

AAC FDK/FHG, is it really the Future of AAC?

Reply #19
I managed to even isolate the artifact on audacity and discovered it's peak at about 12khz frequency. For curiosity I took a look in the spectograph and this is how the artifact looks like:

FFT/window size looks too big for these pictures.

Why is it so? In the horizontal axis you mean? That's not even a 1:1 (pixel per frame) zoom so what's the problem? Well, the pitcture doesn´t really matter that much, the sound does, right? The artifact is pretty clear, When messing with the equalizer in Audacity I found out that the artifact was mostly on the 12khz spectrum and the spectrograph seems to corroborate with that. Would you prefer me sending the files so you could listen to it?

 
SimplePortal 1.0.0 RC1 © 2008-2019