IPB

Welcome Guest ( Log In | Register )

2 Pages V   1 2 >  
Reply to this topicStart new topic
samples where aps or ape has a problem
2Bdecided
post Mar 24 2003, 17:01
Post #1


ReplayGain developer


Group: Developer
Posts: 5139
Joined: 5-November 01
From: Yorkshire, UK
Member No.: 409



I'm sure this has been asked before, but I can't find it, and it's not in the FAQ (it should be - please!)

On which samples is --alt-preset extreme audibly better than --alt-preset standard.
On which samples is --alt-preset insane audibly better than --alt-preset extreme.
On which samples is --alt-preset insane not transparent?

Cheers,
David.
Go to the top of the page
+Quote Post
Wombat
post Mar 24 2003, 18:40
Post #2





Group: Members
Posts: 1038
Joined: 7-October 01
Member No.: 235



There are some,

http://www.hydrogenaudio.org/forums/index....ST&f=16&t=4687&

http://www.hydrogenaudio.org/forums/index....=ST&f=16&t=791&

These i found by just looking into the mp3 - General forum.

You may find more

Wombat
Go to the top of the page
+Quote Post
Pio2001
post Mar 24 2003, 22:11
Post #3


Moderator


Group: Super Moderator
Posts: 3936
Joined: 29-September 01
Member No.: 73



QUOTE (2Bdecided @ Mar 24 2003 - 07:01 PM)
On which samples is --alt-preset insane not transparent?

For me, Badvilbel (by far), and Drone (barely).
Go to the top of the page
+Quote Post
KikeG
post Mar 24 2003, 23:12
Post #4


WinABX developer


Group: Developer
Posts: 1578
Joined: 1-October 01
Member No.: 137



For me, the trumpets.wav sample sounds much more transparent with --api than with --aps or --ape, both of which sound very similar and are not transparent.
Go to the top of the page
+Quote Post
2Bdecided
post Mar 25 2003, 17:25
Post #5


ReplayGain developer


Group: Developer
Posts: 5139
Joined: 5-November 01
From: Yorkshire, UK
Member No.: 409



Thank you all for your responses. I went hunting and listening, and even found another one. To summarise, with --alt-preset standard, I hear the following:


awe22_20sec
http://www.hydrogenaudio.org/forums/index....=ST&f=16&t=2444
ABX 16/16 - the drum beats near the end have added noise

badvilbel
http://www.audio-illumination.org/forums/i...=ST&f=16&t=1059
I (unlike most other people) don't hear a problem - I didn't listen too much because I find the sound painful

birds
http://www.hydrogenaudio.org/forums/index....t=ST&f=16&t=821
ABX 16/16 0 If she's singing "let us beeecome", there's a blip in the left channel on "us"

dogwhistle
http://www.hydrogenaudio.org/forums/index....t=ST&f=16&t=579
I (unlike most other people) don't hear a problem - I didn't listen too much because I find the sound painful ;-)

drone_short
(url to first test?)
ABX 15/16 - sounds noisy during first sound - easy to hear (dropped on ABX due to lack of concentration!)

headache
http://www.hydrogenaudio.org/forums/index....t=ST&f=16&t=423
didn't test

erhu
http://www.hydrogenaudio.org/forums/index....=ST&f=16&t=4687
ABX 14/16 - blip in middle - easy (I almost didn't ABX - not sure what went wrong when I did?!?!)

liebestod
(url to first test?)
ABX 14/16 - difficult (for me) the background is rougher at points, but it's hidden by ambient noise (PC fan mainly) where I'm listening.

sophia2
http://www.hydrogenaudio.org/forums/index....t=ST&f=16&t=543
I (unlike most other people) can't hear a problem - and I did try!

trumpets1
http://www.hydrogenaudio.org/forums/index....=ST&f=16&t=3594
ABX 13/16 - to my ears, the end of the trumpet is less "rough" in the encoded version, but it's very hard to hear

vangelis1
http://www.hydrogenaudio.org/forums/index....=ST&f=16&t=1107
ABX 15/16 - the bass line has a different "character" - it's probably transient smearing, as previously suggested.


Of these, awe32_20sec, birds, erhu and liebestod make me think that, on music I'm likely to encode, there's a very small (but greater than zero) chance of an artefact being encoded. Whether I ever notice it is a different matter!

Cheers,
David.


P.S. Dibrom - does this answer:
http://www.hydrogenaudio.org/forums/index....=ST&f=16&t=3594
still stand? No improvements on the horizon?

P.P.S. in case this thead is read by a newbie as an attack on aps/e/i, please remember that these settings are better than others out there for virtually every audio sample, easy or hard to encode.

P.P.P.S. I think "what samples is aps better than <whatever>" and "what are still problem samples for aps/e/i" should be prominently FAQed! I'm still finding people using FhG VBR! :-(
Go to the top of the page
+Quote Post
Pio2001
post Mar 25 2003, 20:50
Post #6


Moderator


Group: Super Moderator
Posts: 3936
Joined: 29-September 01
Member No.: 73



QUOTE (2Bdecided @ Mar 25 2003 - 07:25 PM)
badvilbel
http://www.audio-illumination.org/forums/i...=ST&f=16&t=1059
I (unlike most other people) don't hear a problem - I didn't listen too much because I find the sound painful

The problem is audible in headphones in a silent environment.
The reverberation after each noise is cut.
The "KRRRRRRsssssss", where "KRRRRRRR" is the noise, and "sssssss" it's decaying echo become
"KRRRRRRss---s-s--s", where - are silences.
Go to the top of the page
+Quote Post
guruboolez
post Mar 25 2003, 21:11
Post #7





Group: Members (Donating)
Posts: 3474
Joined: 7-November 01
From: Strasbourg (France)
Member No.: 420



--alt-preset VBR are suffering with harpsichord. Especially standard, on many harpsichord tracks. You can find a sample here :
http://membres.lycos.fr/guruboolez/

Extreme is better, but not really hard to ABX. --insane is often near-transparency.




Interesting to note : I find Fhg codec bundled with CoolEdit (VBR qual 100 HQ lowpass 20Khz) MUCH better than --alt-preset standard (3.90.2) for the same bitrate. And on a quick audio comparison, I find --ape a bit worse than the same Fhg encoding (but I need to ABX in order to confirm my impression with this sample).
Go to the top of the page
+Quote Post
2Bdecided
post Mar 26 2003, 11:58
Post #8


ReplayGain developer


Group: Developer
Posts: 5139
Joined: 5-November 01
From: Yorkshire, UK
Member No.: 409



guruboolez - thank you - you're right!


With aps I hear...

Fuge
ABX 8/8 - it's noisier behind the notes

Also from your website:

glass_short
ABX 7/8 - there's a "blip" about half way through. It's the same kind of problem as birds and erhu.


also... I know the song, but I've forgotten it, so can you tell me where "Jump" is from please? Artist and Album? Thanks!

btw, why do you .zip some ape files?


Cheers,
David.

P.S. can these samples from other places be added to the HA samples folder please? Specifically, awe32_20sec, and the above two, if it's OK with guruboolez. With sensible all lower case and no spaces filenames in .flac format?

This post has been edited by 2Bdecided: Mar 26 2003, 12:00
Go to the top of the page
+Quote Post
2Bdecided
post Mar 26 2003, 11:59
Post #9


ReplayGain developer


Group: Developer
Posts: 5139
Joined: 5-November 01
From: Yorkshire, UK
Member No.: 409



QUOTE (Pio2001 @ Mar 25 2003 - 07:50 PM)
The problem is audible in headphones in a silent environment.


That would be the problem then - no chance of that! Thanks for letting me know.
Go to the top of the page
+Quote Post
guruboolez
post Mar 26 2003, 12:03
Post #10





Group: Members (Donating)
Posts: 3474
Joined: 7-November 01
From: Strasbourg (France)
Member No.: 420



Jump is a song from Van Halen (1984 album, but not sure).
Funny, but I never notice any artifact with --alt-preset standard on the glass_short sample. I uploaded this sample in order to show the funny behaviour of mppenc at --standard preset (something like 260 kbps for a tonal sample)
I can't put a direct link with a zip or rar file. There's an anti-leech system on Lycos pages.


EDIT : Yes, you're right, a 3.90.2 encoding has the same artifact than with erhu. No problem (or really low) with 3.94.a12. I never noticed it on a larger sample.

This post has been edited by guruboolez: Mar 26 2003, 12:11
Go to the top of the page
+Quote Post
2Bdecided
post Mar 26 2003, 16:57
Post #11


ReplayGain developer


Group: Developer
Posts: 5139
Joined: 5-November 01
From: Yorkshire, UK
Member No.: 409



Reading about the 3.94 alphas, I followed this suggestion from JohnV

http://www.hydrogenaudio.org/forums/index....16&t=6390&st=29


and added -Z to --alt-preset standard in 3.90.2. (I appologise that I'm catching up with this rather late!)

1. It solves the blip problems in erhu, birds, and glass.

2. I think it improves vangelis1 and drone (or else I'm not listening as carefully today - couldn't ABX - but I'm not convinced they're perfect).

3. Fugue ABX 8/9 (didn't notice I'd reached 7/8!) - it's still adding noise - about the same amount

4. It doesn't help awe32_20sec at all. For me the artefacts (the bass drum sound changes) aren't even subtle - I don't need to ABX - I can just get "X" correct without comparison. I've created an even more difficult sample (or obvious problem) by delaying one channel of awe32_20 by 40 samples.


aps -Z bitrate is significantly lower than ape on these samples, and yet the blip in erhu is still obvious with ape. This suggests that, for this sample/problem, ape doesn't deliver the "safety margin" it promisses, and is actually wasting bits without solving the problem. --alt-preset extreme -Z solves the problem, and I think it fixes Fugue too (compared to aps -Z). It doesn't fix awe32_20sec, but it improves it (compared to aps -Z).


Since glass_short wasn't even posted as a problem sample for aps, yet revealed this "blip" problem, which is present in two other (relatively normal) clips, isn't it sensible to suggest --alt-preset standard -Z as the "standard"? Or am I just being too picky?


Cheers,
David.
Go to the top of the page
+Quote Post
2Bdecided
post Mar 27 2003, 11:10
Post #12


ReplayGain developer


Group: Developer
Posts: 5139
Joined: 5-November 01
From: Yorkshire, UK
Member No.: 409



At the risk of being a saddo who just bumps his own threads up... huh.gif

Is there an answer to this question? Or are the people who probably know (Dibrom, JohnV etc etc) more concerned with getting the next stable release right, than checking 3.90.2?

Cheers,
David.
Go to the top of the page
+Quote Post
KikeG
post Mar 27 2003, 11:56
Post #13


WinABX developer


Group: Developer
Posts: 1578
Joined: 1-October 01
Member No.: 137



QUOTE (guruboolez @ Mar 25 2003 - 09:11 PM)
Interesting to note : I find Fhg codec bundled with CoolEdit (VBR qual 100 HQ lowpass 20Khz) MUCH better than --alt-preset standard (3.90.2) for the same bitrate.

According to some tests I performed some time ago, CEP 2.0 Fhg codec, VBR max quality, was noticeabily worse than LAME aps on usual pre-echo test samples, but was better in trumpets1.wav. Still, LAME api was the best on this sample.
Go to the top of the page
+Quote Post
Dibrom
post Mar 27 2003, 12:02
Post #14


Founder


Group: Admin
Posts: 2958
Joined: 26-August 02
From: Nottingham, UK
Member No.: 1



QUOTE (2Bdecided @ Mar 25 2003 - 09:25 AM)
P.S. Dibrom - does this answer:
http://www.hydrogenaudio.org/forums/index....=ST&f=16&t=3594
still stand? No improvements on the horizon?

There will be no further improvements by me if that's what you mean. 3.94 looks like a step in the right direction, but it sounds like there is a lot of tuning left to do, which is always the hardest part. I don't know how things will turn out in the end, but my involvement with LAME has been over for awhile.
Go to the top of the page
+Quote Post
Dibrom
post Mar 27 2003, 12:07
Post #15


Founder


Group: Admin
Posts: 2958
Joined: 26-August 02
From: Nottingham, UK
Member No.: 1



QUOTE (2Bdecided @ Mar 26 2003 - 08:57 AM)
aps -Z bitrate is significantly lower than ape on these samples, and yet the blip in erhu is still obvious with ape. This suggests that, for this sample/problem, ape doesn't deliver the "safety margin" it promisses, and is actually wasting bits without solving the problem. --alt-preset extreme -Z solves the problem, and I think it fixes Fugue too (compared to aps -Z). It doesn't fix awe32_20sec, but it improves it (compared to aps -Z).

Originally, I had planned to modify --alt-preset extreme to use noise shaping 1 but stopped working on LAME before I got around to it. At the time 3.90.2 was released, most of these samples weren't known about and so I believed that I'd finally fixed the artifacts related to noise shaping 2. Most of the problems were solved, but a few remained which were discovered later in the samples you've pointed out.

So, yes, in place of --alt-preset extreme, one should probably use --alt-preset standard -Z instead. FWIW, --alt-preset extreme never really did provide much of an improvement over --alt-preset standard, if any improvement at all. I've never recommended that people use this over --alt-preset standard, and have instead recommened --alt-preset insane if they find --alt-preset standard to be insufficient for some reason. I guess the main reason that I even included an --alt-preset extreme option is because I knew that people would tweak beyond --alt-preset standard even if they couldn't hear the difference anyway, so I figured it'd be best to give them a theoretical improvement rather then have them use some external switches which would negatively impact the --alt-preset behavior.
Go to the top of the page
+Quote Post
2Bdecided
post Mar 27 2003, 13:25
Post #16


ReplayGain developer


Group: Developer
Posts: 5139
Joined: 5-November 01
From: Yorkshire, UK
Member No.: 409



Thank you.
Go to the top of the page
+Quote Post
2Bdecided
post Mar 27 2003, 14:01
Post #17


ReplayGain developer


Group: Developer
Posts: 5139
Joined: 5-November 01
From: Yorkshire, UK
Member No.: 409



To settle aps -Z vs ape -Z vs api in my own mind, I returned to the harpsichord sample "Fugue", because I wasn't convinced by my results yesterday. However, I think I've confirmed them. For me:

aps = aps -Z : the notes are noisy
ape = ape -Z : it's still adding a little noise - just
api is transparent for me


I determined this using ff123's abc/hr comparator. The output was:


ABC/HR Version 0.9b, 30 August 2002
Testname: Fugue Test

1L = D:\audio\lame\apez\wav\Fugue.wav
2L = D:\audio\lame\api\wav\Fugue.wav
3L = D:\audio\lame\aps\wav\Fugue.wav
4L = D:\audio\lame\apsz\wav\Fugue.wav
5R = D:\audio\lame\ape\wav\Fugue.wav

---------------------------------------
General Comments:

---------------------------------------
1L File: D:\audio\lame\apez\wav\Fugue.wav
1L Rating: 2.5
1L Comment:
---------------------------------------
3L File: D:\audio\lame\aps\wav\Fugue.wav
3L Rating: 1.1
3L Comment:
---------------------------------------
4L File: D:\audio\lame\apsz\wav\Fugue.wav
4L Rating: 1.0
4L Comment:
---------------------------------------
5R File: D:\audio\lame\ape\wav\Fugue.wav
5R Rating: 2.4
5R Comment:
---------------------------------------
ABX Results:
Original vs D:\audio\lame\apez\wav\Fugue.wav
8 out of 8, pval = 0.004
Original vs D:\audio\lame\api\wav\Fugue.wav
4 out of 7, pval = 0.500
Original vs D:\audio\lame\aps\wav\Fugue.wav
11 out of 12, pval = 0.003
Original vs D:\audio\lame\apsz\wav\Fugue.wav
8 out of 8, pval = 0.004
Original vs D:\audio\lame\ape\wav\Fugue.wav
7 out of 8, pval = 0.035



(the ratings are relative - none of the files deserves lower than a 4.5 on the true MOS scale)


With awe32_20sec_2 the artefacts are more obvious, so I haven't ABXed today, but the differences are similar to above:

aps = aps -Z : not at all transparent (noise on first two bass drum hits)
ape = ape -Z : much better, but still there
api : even better still - but still very slight problem on first bass drum hit



On Fugue, the size of the difference in quality approximately corresponds to the size of the difference in bitrate.
On awe32_20sec_2, api is much larger than ape, but only a little better. ape is much better than aps, but only a little larger. (exagerated, but you get the point).


It seems there are (at least) two categories of artefact remaining:

one which occurs because of noise shaping 2 and can be removed by -Z

another due to "spiky" waveforms which mp3 just doesn't like, but can be minimised or removed by using the highest possible bitrate --alt-preset insane

maybe yet another due to pre-echo on some (soft?) transients with nspsytune, which can't be solved without another new psymodel.


It was interesting to hear your reason for creating --alt-preset extreme Dibrom. It seems that there is a practical use for the preset though: reducing the audibility of artefacts on the most difficult files, without hitting 320kbps. However, to quote your usual advice, if you really want to encode impossible files "as well as possible", use api or (better still) move to mpc.


The quality goes upwards from aps > aps -Z > ape -Z > api.

ape without -Z doesn't fit into this progression, since it's worse than aps -Z for some samples, but better for others.


Could the FAQ recomended command lines be changed to

--alt-preset insane
--alt-preset extreme -Z
--alt-preset standard -Z
--alt-preset standard

?

Honestly - that's my last question, then I'll drop it!


Cheers,
David.
Go to the top of the page
+Quote Post
2Bdecided
post Mar 28 2003, 11:27
Post #18


ReplayGain developer


Group: Developer
Posts: 5139
Joined: 5-November 01
From: Yorkshire, UK
Member No.: 409



?
Go to the top of the page
+Quote Post
CiTay
post Mar 28 2003, 12:40
Post #19


Administrator


Group: Admin
Posts: 2378
Joined: 22-September 01
Member No.: 3



Hmm, --aps -Z files aren't really that much bigger than --aps. Usually some 1 or 2 kbps increase on average. I also noticed faster encoding with -Z, reproducable. Example 1: With --aps, it goes down to ~4.6x speed and stays there. --aps -Z starts at 6x and levels off at ~5.8x speed. Example 2: --aps speed constantly increasing, reaching 6x towards the end. --aps -Z speed also increasing, but reaching 6.8x. LAME 3.90.2, P4 2.4. The size increase is negligible.

I want to know, is there a consensus about this quality ranking?

--alt-preset insane
--alt-preset extreme -Z
--alt-preset standard -Z
--alt-preset standard

Under these circumstances, i wonder if plain --aps should stay there at all?
Go to the top of the page
+Quote Post
2Bdecided
post Mar 31 2003, 11:16
Post #20


ReplayGain developer


Group: Developer
Posts: 5139
Joined: 5-November 01
From: Yorkshire, UK
Member No.: 409



?

Sorry to bump again, but it's kind of important, isn't it?

D.
Go to the top of the page
+Quote Post
DickD
post Mar 31 2003, 17:14
Post #21





Group: Members
Posts: 265
Joined: 12-January 03
Member No.: 4542



I was searching for info on how -Z worked and found this old thread.

It seems to indicate that there are samples where using -Z to force use of that particular scalefactor can produce problems too, so the specific problem samples you tried, where -Z was no worse, might not be representative of the whole spectrum of music.

Of course, in the present thread, Dibrom has noted that certain samples (like those you tested) weren't known about then, so perhaps his opinion is now revised.

--alt-preset standard is well tested and lacks transparency only on rare cases. I don't think --alt-preset standard -Z has been tested anywhere near as much and shown to be OK, so I won't be using it, but I might try it if I notice a specific problem with APS (which won't be too often, now I mostly encode in Musepack mppenc --quality 5 --xlevel).
Go to the top of the page
+Quote Post
KikeG
post Apr 1 2003, 10:05
Post #22


WinABX developer


Group: Developer
Posts: 1578
Joined: 1-October 01
Member No.: 137



IIRC, JohnV said that there's no problem in using -Z always, other than a bitrate increase.
Go to the top of the page
+Quote Post
2Bdecided
post Apr 1 2003, 10:24
Post #23


ReplayGain developer


Group: Developer
Posts: 5139
Joined: 5-November 01
From: Yorkshire, UK
Member No.: 409



JohnV did say exactly that, but Dibrom said "and by specifying -Z you may end up degrading quality in cases as well."

I can't figure out if he had any specific for saying this, because it came in one of the (at the time) very frequent replies which essentially said "Look, don't worry about the switches which are in --alt-preset standard, because there are things which can't be accessed from the command line; and don't try to improve it by adding other switches, because it's already as good as it can get, and more importantly, you'll almost certainly make it worse by trying to 'tweak' it".

So maybe he had a very specific problem in mind, or maybe he was just trying to stop people messing with the --alt-presets.

Dibrom - please tell us!

Cheers,
David.
Go to the top of the page
+Quote Post
2Bdecided
post Apr 3 2003, 15:52
Post #24


ReplayGain developer


Group: Developer
Posts: 5139
Joined: 5-November 01
From: Yorkshire, UK
Member No.: 409



?



With the many tests samples I've been trying, I am slowing coming to an even more contraversial conclusion... r3mix is often better than --aps!!!


Sorry - only joking! I should have tried this one on April 1. wink.gif



Seriously though - I have found something which no-one here has ever said, probably because no one uses 128kbps CBR mp3: --alt-preset cbr 128 is often worse than -b 128 -h (i.e. the lame default). ap cbr 128 is probably better in the majority of cases (though I wouldn't bet my life on it), but there are many many samples where -b 128 -h is actually better than --alt-preset cbr 128.

This isn't a late April fool - try the samples listed in this thread for starters. Much of the time neither command line is transparent, so to some extent it depends on which artefacts you find most annoying; but there are some cases where lame default beats the "tuned" line hands down.


Just thought it was worth mentioning. I've already PMed ff123, who helped develop the command line behind --alt-preset cbr 128. Maybe the fact that it's worth trying other things (even other encoders) at 128kbps should be in the HA FAQ - and/or a subject for further lame tuning?


Anyway, what really interests me is the unanswered aps/e -Z issue - can adding -Z ever decrease quality? Please tell!

Cheers,
David.
Go to the top of the page
+Quote Post
JohnV
post Apr 3 2003, 17:57
Post #25





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



It has been said many times that Gpsycho has often for example better pre-echo control than plain vanilla nspsytune preset, so I don't know if that's anything new. Many times it has been said that the non-codelevel tweaked presets aren't necessarely always performing better than some other switches with a bunch of samples. The --alt-preset cbr 128 also suffers (although prolly not quite similarly) from the use of noise shaping type 2. Try the samples listed in this thread using --alt-preset cbr 128 -Z although this may not be ideal overall... And since GPsycho uses noise shaping type 1 by default, it performs very well with those samples which fail with ns-type2...

gpsycho plain vanilla -b128 -h is indeed not bad for 128kbps. But if you use abr, I think the advantage will turn more clearly to nspsytune considering overall quality aspects.

And no, in vbr coding I don't believe that using noise shaping 1 (-Z with APS/APE) will lead to worse quality, but it definitely leads to better quality in some cases. Lower quality with noise shaping 1 is definitely possible with lower bitrate cbr though.


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

2 Pages V   1 2 >
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: 21st September 2014 - 09:17