IPB

Welcome Guest ( Log In | Register )

2 Pages V   1 2 >  
Reply to this topicStart new topic
Problem with --standard, up to --quality 10, 1.14 & 1.15r confrontation inside...
guruboolez
post Aug 7 2003, 16:22
Post #1





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



Bored by some recent discussion about the relevance of high settings with musepack encoder, I tried to prove to myself and to other people that:
- finding slight problems samples for mpc –quality 5 isn’t too difficult (even if it needed a lot of concentration to ABX problem)
- these small problems will disappear with a higher bitrate.

I suspect since a long time that mpc isn’t clear with cymbals. I have a lot of metal music encoded with –standard, and I often perceive some suspicious distortions with this instrument. Unfortunately, I haven’t the original for ABX test. I tested some long times ago on some cymbals sample I founded myself, but conclusions were clean for mpc.
Today, I decided to try again, more seriously, and therefore I searched for a good cymbal sample in my entire library. A name quickly came into my mind : Richard Wagner… I found some sharp cymbals in his famous hit, popularized by Francis Ford Coppola. I cut some seconds, and encode them with both 1.14 and 1.15r (I’m not convinced that 1.15r is better than 1.14, and I also suspected 1.14 to be stronger on most ‘regular’ samples than 1.15r). –xlevel encoding was always enabled, to avoid clipping distortion.

First, I ABXed –standard encodings. On the first cymbal attack, I was disappointed. Though I didn’t listen very carefully, I expected a more obvious difference. I jumped few seconds, and ABXed another part of the file (cymbals too) : from second 10.00 to 11.5. Results are here :

STANDARD
(first trial)
1.14 : 4.2 - 12/16
1.15r : 4.5 - 12/16
1.14 vs 1.15r : 9/16
(second trial)
1.14 : 4.2 - 13/16
1.15r : 4.5* - 12/16 (* mixed up original and encoded)
1.14 vs 1.15r : 2/7

Though I rated two times 1.14 at a higher level than 1.15r, I wasn’t able to distinguish them on a direct opposition. Most important thing : --standard profile is ABXable on this small part/high cymbal attack.

Let see if –extreme will correct the problem :

EXTREME
1.14 : 4.8 - 8/16
1.15r : 4.3 - 16/16
1.14 vs 1.15r : 13/16 (pval = 0.011)

I didn’t perform a second test at this setting. Result were obvious, confirmed by the good ABX score I had with 1.15r (16/16 !) and with direct opposition (13/16). 1.15r was significantly worse, due to an annoying whistle I heard with evidence. On the other side, 1.14 was near transparent. On my quick test, I couldn’t ABXed it. I was happy, and decided to perform a last test at –insane…

This post has been edited by guruboolez: Aug 7 2003, 16:34
Go to the top of the page
+Quote Post
guruboolez
post Aug 7 2003, 16:29
Post #2





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



INSANE
1.14 : 4.4 - 12/16 (pval = 0.038)
1.15r : 4.8 - 8/16
1.14 vs 1.15r : 18/32 (pval = 0.298)
1.15r seemed to be better with –insane, but I couldn’t proof it with direct opposition. I tried a second time.
(second trial)
1.14 : 4.7 – 14/20 (pval = 0.058)
1.15r : 4.7 – 15/20 (pval = 0.021)
1.14 vs 1.15r : 15/20 (pval = 0.021)
This time I was sure : 1.14 was more distorted than 1.15r. The total opposite of –extreme performances ?!
I decided to go with braindead, encode the file, but I accidentally forgot to change config file. I tested insane a third time :
(third trial)
1.14 : 4.6 – 13/16 (pval = 0.011)
1.15r : 4.8 – 10/16 (pval = 0.227)
1.14 vs 1.15r : 13/16 (pval = 0.011)
Again, 1.14 sound less natural and more distorted ; confirmed on ABX direct confrontation… Now, braindead.

BRAINDEAD
1.14 : 4.8 – 14/16 (pval = 0.002)
1.15r : 4.9 – 13/16 (pval = 0.011)
1.14 vs 1.15r : 16/29* (pval = 0.356) * I expected better result after a bad beginning – it wasn’t the case…
I must say that I was shocked by these results. I tried to prove that higher bitrate are sufficient to correct artifacts, I now, I came to the conclusion that –quality 8 wasn’t transparent on a sample I chose by feeling !! I followed my investigations, sad for my favorite codec, but happy… happy to fight against a monster, and to progressively win the match.

This post has been edited by guruboolez: Aug 7 2003, 16:30
Go to the top of the page
+Quote Post
guruboolez
post Aug 7 2003, 16:30
Post #3





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



QUALITY 9
1.14 : no notation – 12/16 (pval = 0.038)
1.15r : no notation – 5/16 (pval = 0.962)
1.14 vs 1.15r : 6/16 (pval = 0.895)
Best and conclusive ABX score for 1.14 – 1.15r was transparent. But I began ABX test with 1.15r, and 1.14 that came after may have suffered from this warm-up. So, if direct confrontation wasn’t conclusive, I have ABXed –quality 9 with 1.14 encoder, on a sample that didn’t sound very critical : very annoying for me…

QUALITY 10
1.14 : 4.8 – 10/16 (pval = 0.227)
1.15r : no notation – 8/16 (pval = 0.598)
1.14 vs 1.15r : no test
(second trial)
1.14 : 4.8* – 14/20 (pval = 0.058) *confused
1.15r : 4.8 – 14/20 (pval = 0.058)
1.14 vs 1.15r : 2/6 (I quickly gave up)
Conclusive ABX results for both encoders. There is a small problem (for me, it’s distortion : can’t be more precise), even at high bitrate (330 kbps). And at this moment of the test, I didn’t had to be very attentive to hear this problem….



I’m forced to agree, now, with people claiming that high bitrate encodings aren’t really worth. And the trust I usually had for my whole --insane encodings just exploded on my face… Now, on the first suspicious sound I will hear, I probably will accuse mpc for this…
Nevertheless, I can’t explain my performance. Has the scorching heat a positive impact on my hearing abilities ? Is it because I listened very few music since five weeks ? Had my ears progress, and sensitive as never there were for mpc artifacts ? Maybe other people will confirm the slight degradation I heard (please, try on this small part !). At least, the strange phenomenon I heard against 1.15r encoder (the one 16/16 I obtained during the whole test).

Sample is now uploading…
Go to the top of the page
+Quote Post
guruboolez
post Aug 7 2003, 16:37
Post #4





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



Sorry for this flooding. Forum is broken...
I can't edit my post in order to make them more pleasant, without bugging the text...

Sample is here :

http://membres.lycos.fr/guruboolez/AUDIO/s...ries_short.flac
Go to the top of the page
+Quote Post
Dologan
post Aug 7 2003, 17:38
Post #5





Group: Members (Donating)
Posts: 478
Joined: 22-November 01
From: United Kingdom
Member No.: 519



No, can't ABX and man, I tried! Of course I am no golden-eared such as yourself, but for me no difference. I hope one does hear it, though.
Still, the conclusions of this will, however, seriously support the view that advocates: if -q 5 fails, going up won't help, so just stay at -q 5.

~Dologan

This post has been edited by dologan: Aug 7 2003, 17:39
Go to the top of the page
+Quote Post
guruboolez
post Aug 7 2003, 17:44
Post #6





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



Have you tried the specific part I suggest : 10.00 - 11.50 ? With an headphone ?
Anyway, thanks for testing.
Go to the top of the page
+Quote Post
music_man_mpc
post Aug 7 2003, 17:48
Post #7





Group: Members (Donating)
Posts: 707
Joined: 20-July 03
From: Canada
Member No.: 7895



This is indeed most distressing, but I am glad that you found this problem. In the Post 128Kbit test ff123 said:

QUOTE
Several people have very good high frequency hearing, and can hear the lowpassing of MPC and AAC (both around 16 kHz). One person (gecko) seems to find this to be a significant defect.


Is it possible that this is what you are hearing and that raising the lowpass filter *might* help? It might explain why there is little or no quarrelation between the different --quality settings tried.


--------------------
gentoo ~amd64 + layman | ncmpcpp/mpd | wavpack + vorbis + lame
Go to the top of the page
+Quote Post
guruboolez
post Aug 7 2003, 17:53
Post #8





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



QUOTE (music_man_mpc @ Aug 7 2003, 05:48 PM)
Is it possible that this is what you are hearing and that raising the lowpass filter *might* help?  It might explain why there is little or no quarrelation between the different --quality settings tried.

mpc lowpassing is near inexistant at high setting. It's not the problem.
Note that the original file doesn't reach 22 khz : recording is old (Bayreuth, 1967).
Go to the top of the page
+Quote Post
Dologan
post Aug 7 2003, 17:56
Post #9





Group: Members (Donating)
Posts: 478
Joined: 22-November 01
From: United Kingdom
Member No.: 519



QUOTE (guruboolez @ Aug 7 2003, 10:44 AM)
Have you tried the specific part I suggest : 10.00 - 11.50 ? With an headphone ?
Anyway, thanks for testing.

Yeah. That very part and of course with headphones. The equipment is not high-end (just my laptop sound card and a pair of Sennheiser PX100s), but that's the best I've got right now.

This post has been edited by dologan: Aug 7 2003, 17:59
Go to the top of the page
+Quote Post
mithrandir
post Aug 7 2003, 18:18
Post #10





Group: Members
Posts: 669
Joined: 15-January 02
From: SE Pennsylvania
Member No.: 1032



I don't think there is any need to be shocked since, again, MPC is lossy. Artifacts are theoretically (and often pragmatically) possible in any situation regardless of quality setting used. Remember also that SV7 creates a few relatively minor audible problems due to its design that you can't remove using bitrate alone. SV8 is supposed to address these problems (though no guarantees it will be problem-free itself), assuming an SV8-compliant mppenc is ever released.
Go to the top of the page
+Quote Post
guruboolez
post Aug 7 2003, 18:30
Post #11





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



QUOTE (mithrandir @ Aug 7 2003, 06:18 PM)
I don't think there is any need to be shocked since, again, MPC is lossy. Artifacts are theoretically (and often pragmatically) possible in any situation regardless of quality setting used.

Yes, but don't forget the purpose of this test, explained at the beginning : I tried to prove that artifacts will be removed with bitrate inflation, and if standard isn't sufficient for that, insane will. Now, i'm bloody annoyed by the result :'(
I had a great faith in mpc format. I'm disappointed... and surprised to defeat --quality 10 (on one sample, but very common sound).

Fortunately, artifact is limited, and not shocking. And for most of my 10000 mpc encoding, I have the original, so I can rerip them if needed... in lossless wink.gif
Go to the top of the page
+Quote Post
NumLOCK
post Aug 7 2003, 18:53
Post #12


Neutrino G-RSA developer


Group: Developer
Posts: 852
Joined: 8-May 02
From: Geneva
Member No.: 2002



It confirms a bit what I've been thinking:

- every time you switch for a higher setting, more margin is added to the psymodel's parameters.
- this means, even at highest settings the actual quality margin (inside the output file) is always chosen by the psymodel. So if the psymodel badly fails at a certain place, and you step up the quality profile, there's little reason (from encoder's point of view) to allocate more bits there.

The only way to remove artifacts for sure, would be to force, for example, an additionnal bit in each of the smallest time units. Downside: while helping artifacts, this would waste lotal quality margin (inside the output file) is always chosen by the psymodel. So if the psymodel badly fails at a certain place, and you step up the quality profile, there's little reason (from encoder's point of view) to allocate more bits there.

The only way to remove artifacts for sure, would be to force, for example, an additionnal bit in each of the smallest time units. Downside: while helping artifacts, this would waste lots of bits..


--------------------
Try Leeloo Chat at http://leeloo.webhop.net
Go to the top of the page
+Quote Post
Gecko
post Aug 7 2003, 19:31
Post #13





Group: Members
Posts: 948
Joined: 15-December 01
From: Germany
Member No.: 662



Nice one! I may have missed it, but does the artifact get better with increasing bitrates? I should consider downgrading from my current quality 5.5 to pure quality 5. To hell with 0.5 quality points of warm fuzzy feeling!
Go to the top of the page
+Quote Post
Dologan
post Aug 7 2003, 19:36
Post #14





Group: Members (Donating)
Posts: 478
Joined: 22-November 01
From: United Kingdom
Member No.: 519



@Guruboolez: Have you tried using a Wavpack/Optimfrog lossy file at a healthy bitrate (448 or 512 kbps)? Maybe it will do the trick for you without having to cope with the filesizes of lossless. Is a Wavpack/Optimfrog lossy transparent to your ears at this bitrate?
Go to the top of the page
+Quote Post
Dologan
post Aug 7 2003, 19:46
Post #15





Group: Members (Donating)
Posts: 478
Joined: 22-November 01
From: United Kingdom
Member No.: 519



Well, judging from his marks, apparently version 1.14 improves from standard to extreme and stalls more or less there at 4.8. Version 1.15j appears better at the beginning, but reaches its maximum at insane, also 4.8.
This is obviously not enough data to be able to reach a fair conclusion, but if one may be made, it would be that it makes sense to go up to extreme, but not further, unless you suffer from some byte-wasting compulsion that could be classified as "insanity" or "braindeath".
Go to the top of the page
+Quote Post
guruboolez
post Aug 7 2003, 20:16
Post #16





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



I listen to classical music only. Lossless encoding are a bit superior to 40 % (~600 kbps). Therefore, lossy encodings at 450 kbps don't appear to me as a pertinent choice.
I didn't try dualstream or wavepack lossy. Except a very slight level of noise, I don't expect any distorsion with this kind of encoder.

And don't put too much attention on notation : 4.6, 4.8... it doesn't mean anything else that I heard a small difference (and I'm not always sure, before ABX test).
Go to the top of the page
+Quote Post
indybrett
post Aug 7 2003, 20:21
Post #17





Group: Members (Donating)
Posts: 1352
Joined: 4-March 02
From: Indianapolis, IN
Member No.: 1440



Stand by for the flood of "that's why I use lossless" posts wacko.gif


--------------------
flac>fb2k>kernel streaming>audiophile 2496>magni>dt990 pro
Go to the top of the page
+Quote Post
Pio2001
post Aug 7 2003, 22:10
Post #18


Moderator


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



mppenc 1.14 --quality 10 --xlevel
Seconds 8.2 to 8.9
ABX 2/5

Failed.
Go to the top of the page
+Quote Post
guruboolez
post Aug 7 2003, 22:17
Post #19





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



Did you ABXed this part with --quality 10 first ? I suggest to warm up your ears with --quality 5, or even --quality 4.

I tried to ABX the same file some minutes ago, directly at --insane. After a dozen total failure, I began to hear the distorsions again.
Go to the top of the page
+Quote Post
Pio2001
post Aug 7 2003, 23:06
Post #20


Moderator


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



I tried quality 4 first, but could not ABX at quality 10 then. I'll try the part you suggested later (10-11 s)
Go to the top of the page
+Quote Post
phong
post Aug 8 2003, 03:11
Post #21





Group: Members
Posts: 346
Joined: 7-July 03
From: 15 & Ryan
Member No.: 7619



Hmm, this one's beyond my ability...

Just for kicks, I started digging in my collection to see if I could find any samples that make trouble for mpc --standard... OK, so I started with something easier; finding a track that makes trouble for lame --alt-preset standard. smile.gif Found one in the first 10 seconds of the first track I tried!
Aphex Twin - Bucephalus Bouncing Ball (Come to Daddy EP).

Very easy. ABXed 10/10 during a rainstorm (no joke!) Tried --alt-preset insane, still 10/10 (rainstorm was over, but still had a pretty noisy computer). Tried oggenc -q6 (gt3b1) - tougher, but definately still noticeable. I willing to accept that this track is a bit of a pathalogical case though. :-)

Mpc... Well, I'm still working on that.


--------------------
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/
Go to the top of the page
+Quote Post
TrNSZ
post Aug 8 2003, 09:33
Post #22





Group: Developer
Posts: 717
Joined: 25-September 01
Member No.: 20



ABX of 10-11.5 was trivial even at Q10 - 13/14. But, I believe these samples are still special cases, and bringing them up will do nothing but help make improvements.
Go to the top of the page
+Quote Post
guruboolez
post Aug 8 2003, 09:53
Post #23





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



QUOTE (TrNSZ @ Aug 8 2003, 09:33 AM)
ABX of 10-11.5 was trivial even at Q10 - 13/14.  But, I believe these samples are still special cases, and bringing them up will do nothing but help make improvements.

Glad to see a positive report. How do you describe the problem ? I can't be more precise than 'distortion'.
Why would this sample be a special case ? I remind that I didn't heard something wrong before I started ABX tests : I choose this sample because I suspected MPC to have problem with instruments like cymbals. Maybe had I luck, and catch for a first trial a very good sample. But can't we suppose that most of sharp and powerful cymbals attacks are under menace ? However, I must admit that the presence of strings in the background is maybe playing a role in this artifact ; that's maybe why previous cymbals aren't affected so much (or at all), although they sound quite the same. Cymbals, alone, are maybe well reproduced by mpc...

What do you think about it ?
Go to the top of the page
+Quote Post
Pio2001
post Aug 8 2003, 12:04
Post #24


Moderator


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



The Smashing Pumpkin's sample of Xerophase was also chosen quite randomly, and you also found harpsichord to have problems most of the time. It seems that these samples are not pathological but stand for a common behaviour of MPC.

The difference with MP3 is the nature and strenght of the artifacts. They are more subtle in nature, in last resort, I reduced the Smashing pumpkin "helicopter" artifact as a very quiet treble noise. That's how I could ABX it best. MP3 artifacts seems stronger to me, and different : pre-echo, drop outs...
Also, MP3 artifacts seem to occur only on pathological samples, while MPC artifacts seem to affect a wide variety of treble sounds.

QUOTE
Now, on the first suspicious sound I will hear, I probably will accuse mpc for this

That's what I'm constantly doing, but most of the times, I must admit that it's the original that sounds like an MPC artifact, and the compressed file sounds exactly the same. I tried parts of Kovenant, Arcana, MacTatus, that seems to have nasty MPC artifacts, but no, the originals sound like this. Astral Projection sounds exactly the same, exept that the effect is amplified after MPC encoding.
I really have to try to ABX in order to distinguish between real and fake MPC artifacts.
That doesn't turn MPC artifact unimportant for me. Even if this kind of sound can be found in original recordings, I don't like the sound of it.
Go to the top of the page
+Quote Post
guruboolez
post Aug 8 2003, 12:29
Post #25





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



QUOTE (Pio2001 @ Aug 8 2003, 12:04 PM)
QUOTE
Now, on the first suspicious sound I will hear, I probably will accuse mpc for this

That's what I'm constantly doing, but most of the times, I must admit that it's the original that sounds like an MPC artifact, and the compressed file sounds exactly the same. I tried parts of Kovenant, Arcana, MacTatus, that seems to have nasty MPC artifacts, but no, the originals sound like this. Astral Projection sounds exactly the same, exept that the effect is amplified after MPC encoding.
I really have to try to ABX in order to distinguish between real and fake MPC artifacts.
That doesn't turn MPC artifact unimportant for me. Even if this kind of sound can be found in original recordings, I don't like the sound of it.

The first months I decided to switch from lame to mpc, I had very bad feelings : heavy distortions on most recording. I was litteraly crazy about potential artifacts of this unknow format - absolutely no trust on its quality on classical music... I ABXed many of the parts that sounded really odd to me, ut all were negative, even with --standard profile (I encoded -and 'heard' distortions- with insane !). I was so paranoïac at this period that I 'felt' artifacts and compressed sound... on original CD !
After so many ABX failure, I went to the conclusion that placebo effect was really strong, and that I was victim of a great audiophile paranoïa. After these tests, I decided to put no more attention on potential distortions. I know that original CD, or even instruments (baroque instruments are very and naturally distorted - harpsichord sound is based on distortion), are the only responsible of what I heard. Sometimes, I tried to ABX little part that really annoyed me, but with no succes (but maybe due to unwillingness).

Problem : the trust I hardly developped with time is now broken with this test. Of course, I had to find some others samples, ABXable up to --insane or higher profile...

Nevertheless, you're right : mpc artifacts are really subtle.
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 December 2014 - 12:50