Skip to main content

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: Personal evaluation at ~130..135 kbps, 200 samples (Read 156298 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Personal evaluation at ~130..135 kbps, 200 samples

Reply #75
Quote
Nevertheless, there are some obvious regression, occuring with different kind of samples (like harpsichord, organ and other tonal signal). That's why I'm disappointing to see the new encoder not following some progress revealed last year (with classical at least). But it's much much better than 2003 encoders
[a href="index.php?act=findpost&pid=342508"][{POST_SNAPBACK}][/a]


I believe this is an issue of specific listener preference/sensitivity + genre, and that your conclusion is not generally applicable (and in fact, most people will experience the opposite). That's why, for example, we never really recommended fast mode, despite you saying it was much better.

It may be possible to "fix" it for both groups and we have made some progress on this in the meantime.

Personal evaluation at ~130..135 kbps, 200 samples

Reply #76
Quote
I'd really appreciate if you have time to answer any of these questions.
[a href="index.php?act=findpost&pid=342599"][{POST_SNAPBACK}][/a]

I have the bad feeling that someone from CIA or KGB is questionning me...

Do you play an acoustical unamplified instrument? If so, which?
Unfortunately, now. If I had, it would be harpsichord, gamba or cello.

Do you sing or have you been taugh singing as a trained skill?
No. I'm fond of baroque music (especially opera), and it's too late for me to be a castrato (it's also forbidden).

Have you trained your pitch listening skills consciously (relative or absolute)?
My what?

Which one is your leading ear (more accurate in tonality estimation)? Do you use both ears throughout all the tests or sometimes just one ear?
A leading ear? I'm right-handed, but my hearing seems ambidextrous.

Have you consciously trained by listening to test sample sets (known errors in perceptual coding)? Or just listened to a lot of lossy encoder output?
I don't think that listening to lossy encodings would help to pass a listening test. Otherwise there would be millions golden-ears on this planet. In my case, my training ground is ABC/HR and not foobar2000 or winamp.

At which volume levels you usually listen to (soft, medium, loud)? Or do you vary volume throughout the test?
It's hard to tell. Some people are convinced to listen at moderate volume when they're sharing their music with the full train with a simple pair of earbuds. I'd say "medium". Sometimes very loud (enthusiastic mood), sometimes very quiet. During listening test, it's "medium" (sometimes louder, sometimes quieter).

Have you ever listened to any of the encoders via loudspeakers (not for tonality issues or small artifacts, but for soundstaging issues)?
No.

At what time of the day do you you usually test?
As soon as I can (200 samples: no time to loose).

How long tests can you do before you have to stop or find that you can't make out the differences anymore (number of repeats or duration roughly in minutes)?
It depends. Some samples are less exhausting than others. The enthusisasm for the music plays a big role also. No need to precise that I've more energy for my own music than for something I don't like.
I'm often taking a break; some of them a very short (they may happen during ABC or ABX phases, and not necessary between two samples). I'm also used to stop few minutes between two samples. After 5...15 samples (without ABXing), I must stop longer.

Do you use a lot of quick back-and-forth A<->B switching or do you just listen to sample A in full, then sample B in full and then decide?
It depends on the tested bitrate. Here, degradations are often obvious. That's why I could often pull down the slider within two or three seconds for each sample. I rarely hit the X button. I'm working in few steps:
- first listening pass: I'm trying to quickly unmask each encoding (listening to all A-B A-B A-B A-B, first seconds only). I'm giving an approximate notation according to the level of annoyance.
- second listening pass to pull down the remaining sliders and to adjust (still vaguely) the sliders.
- if encoders are still unmasked, I'm changing the playback range and listen carefully. (again, I'm trying to increase the accuracy of the ranking at this stage).

Then begins the second big step, and it's the most longer: time for me to give an accurate mark and most dramatic point, to decide for a hierarchy. People which are not used to perform listening tests might not understand why it's so hard, but I can swear you that grading different form of distortions is anything but obvious. Sometimes the hierarchy is not debatable (it happens with discriminative samples), but most often it's problematic. Another point: each contenders makes this step longer and tedious. That's why I'm  when I see people asking to test 5, 6 or sometimes 7 files in collective tests. I suppose that some of them not realizes the amount of work that each additional competitor implies.


What are the top 3 most important things for yourself when you do a listening test?
- Coffee
- Tea
- listening to a full composition instead of 10 seconds samples from time to time (this, and all other kind of breaks)

Less materially: it's very important that the test makes sense. If I have the feeling that one of the choosen encoder or setting is not the good one, it ruins the motivation. That's why it's very important to know the average bitrate of the VBR preset, or to know if you're testing the good implementation or setting. Otherwise, the test becomes useless and is just a waste of time.


PS I just smile at the thought of what kind of accuracy you'd reach if you had a proper pair of electrostatic headphones at your use (very low distortion, very good tonality), a powerful noiseless/distortion-free amp and a good sound card.

Results won't be different I suppose. They won't change the accuracy at least, simply because all grading can't be accurate by themselves (it's often 2.0, 2.5, 3.0, 3.5... and this won't change with better or poorer hardware )
A better headphone might help me to find additionnal subtle distortions, but to make progress training is always better than equipment.

Personal evaluation at ~130..135 kbps, 200 samples

Reply #77
Quote
I believe this is an issue of specific listener preference/sensitivity + genre, and that your conclusion is not generally applicable (and in fact, most people will experience the opposite).
[{POST_SNAPBACK}][/a]


Belief and facts in the same time 
How do you know that "most people will experience the opposite"?
I don't think that what I heard is specific to my hearing. I can't prrove it, but I can try to experiment this, and post samples encoded with 3.2.0.15 and 4.2.1.0 and illustrating what I consider as major regression.
• [a href="http://foobar2000.net/divers/temp/S22.zip]S22: organ[/url] (622 Kb)
S30: accordion (428 Kb)
S44: bagpipe (900 Kb) - all archives contains the encoded files, and all are named (I'm not trying to fool nor to play with people).

The quality is very different; the bitrate too. The previous "fast" encoder which I considered as amazingly good was used to spend a lot of bits to maintain quality with very tonal signal (bitrate is very high, but on very short moments only). The current new encoder reacts much more like the previous "high" encoder, and with poorer quality sometimes!
Extreme case is S44:
aacenc3 'high' = 134 kbps
aacenc3 'fast' = 302 kbps
aacenc4 'high' = 113 kbps

I also put iTunes encoding for comparison: there's no bloated bitrate, and quality is in my opinion near transparency.

[span style='font-size:12pt;line-height:100%']It would be nice to see other people downloaded these three examples, and to give their opinion. I'd like to know if my hearing is very far from normality, or if other people agree with me. It would be hard to believe that what I hear are issues that "are not generally applicable". Thanks[/span]

Original files are also here: S22, S30 & S44


Quote
That's why, for example, we never really recommended fast mode, despite you saying it was much better.

I said it, and Ivan and JohnV confirmed it. The "fast" encoder was supposed to correct real issues occuring with the "old/high" encoder and mostly audible with classical. And this new encoder was announced to be the successor of the old one when some problems (there were never detailed, even when I asked for them) will be fixed. Now it seems that the successor don't even include the enhancement of the previous experimental encoder 

Personal evaluation at ~130..135 kbps, 200 samples

Reply #78
@guruboolez,

There will be an update for the next multiformat listening test that I hope correct the issues you found.

I hope you will find some time to test the upcoming version.

Personal evaluation at ~130..135 kbps, 200 samples

Reply #79
Quote
It would be hard to believe that what I hear are issues that "are not generally applicable".
[a href="index.php?act=findpost&pid=343105"][{POST_SNAPBACK}][/a]


What part of "genre" in my previous post didn't sink in?

You're now intending to 'prove' that point by posting 3 classical samples where one encoder spends two or three times as much bits as the other, and test if people can hear the difference? Eh?

I do not believe Nero 4.2.1.0 "high" is a regression over 3.2.0.15 "fast" mode.

I can believe that for classical it's especially bad for you.

I do not believe that will be generally true.

Personal evaluation at ~130..135 kbps, 200 samples

Reply #80
Quote
You're now intending to 'prove' that point by posting 3 classical samples where one encoder spends twice as much bits as the other, and test if people can hear the difference? Eh?
[a href="index.php?act=findpost&pid=343112"][{POST_SNAPBACK}][/a]


First: I'm not trying to prove anything. Read what I said: "I can't prrove it, but I can try to experiment this...".
And I believe that posting three samples as "illustration" (I'm quoting myself) is surely better than constantly making claims about "most people hear" without any elements of proof.

Second: try to answer correctly please. You perfectly know that the bitrate you get on short samples is not a problem. The previous -internet mode and the current one both lead to similar bitrate (~130 kbps) and if the previous encoder was clever enough to allocate 300 kbps to maintain the quality, then the current one has issue.

Quote
I do not believe Nero 4.2.1.0 "high" is a regression over 3.2.0.15 "fast" mode.

That's fine. I personaly don't believe anything on this subject: I haven't compared directly the previous encoder with the new one. But I know that the previous "fast" encoder handled very well most situations I've tested, and that I can't say the same thing for the new one. With some samples, quality really seems to me unworthy of a modern AAC implementation.
What do you think about the samples I've posted as illustration?



P.S. enable on your side the option "this post as been edited..."

Personal evaluation at ~130..135 kbps, 200 samples

Reply #81
Just posting my initial impressions, in response to guruboolez' personal invitation to me.  I've had a listen to S30 and will post further results on the other sample when my ears aren't as tired

NB:  I'm far from being golden-eared.

Sample S30

Code: [Select]
-------------------------------------
WinABX v0.23 test report
11/19/2005 22:57:23

A file: C:\Documents and Settings\stephen\Desktop\S30\001 S30 [4.2.1.0 VBR internet high].wav
B file: C:\Documents and Settings\stephen\Desktop\S30\001 S30 [iTunes v6.0.0.18, QuickTime 7.0.3].wav

22:58:16    1/1  p=50.0%
22:59:56    2/2  p=25.0%
23:00:10    3/3  p=12.5%
23:00:32    4/4  p= 6.2%
23:01:08    5/5  p= 3.1%
23:02:07    6/6  p= 1.6%
23:03:31    7/7  p= 0.8%
23:04:03    8/8  p= 0.4%
23:04:23    9/9  p= 0.2%
23:04:50  10/10  p< 0.1%
23:05:08  11/11  p< 0.1%
23:05:41  12/12  p< 0.1%
23:05:42  test finished

-------------------------------------
WinABX v0.23 test report
11/19/2005 23:08:11

A file: C:\Documents and Settings\stephen\Desktop\S30\001 S30_OTHERS_Accordion_A.wav
B file: C:\Documents and Settings\stephen\Desktop\S30\001 S30 [4.2.1.0 VBR internet high].wav

23:09:13    1/1  p=50.0%
23:09:31    2/2  p=25.0%
23:10:32    3/3  p=12.5%
23:11:45    4/4  p= 6.2%
23:11:57    5/5  p= 3.1%
23:13:20    6/6  p= 1.6%
23:13:49    7/7  p= 0.8%
23:14:06    8/8  p= 0.4%
23:14:18    9/9  p= 0.2%
23:14:37  10/10  p< 0.1%
23:14:54  11/11  p< 0.1%
23:15:05  12/12  p< 0.1%
23:15:07  test finished

-------------------------------------
WinABX v0.23 test report
11/19/2005 23:21:05

A file: C:\Documents and Settings\stephen\Desktop\S30\001 S30 [3.2.0.15 VBR internet fast].wav
B file: C:\Documents and Settings\stephen\Desktop\S30\001 S30 [4.2.1.0 VBR internet high].wav

23:21:36    1/1  p=50.0%
23:21:57    2/2  p=25.0%
23:22:37    3/3  p=12.5%
23:23:02    4/4  p= 6.2%
23:23:18    5/5  p= 3.1%
23:23:35    6/6  p= 1.6%
23:24:00    7/7  p= 0.8%
23:24:35    8/8  p= 0.4%
23:25:07    9/9  p= 0.2%
23:25:26  10/10  p< 0.1%
23:25:42  11/11  p< 0.1%
23:26:01  12/12  p< 0.1%
23:26:04  test finished


One ABX test I didn't include above is between iTunes AAC and the original. Basically I failed it and couldn't tell the difference without guessing, so I gave up.

Basically, there seems to be a sharp amplification of the change in notes between 1 and 2 seconds in the "4.2.1.0 VBR internet high" encoded file that was very apparent, which I didn't hear in either the iTunes, "3.2.0.15 VBR internet fast" sample, or the original sample, hence the perfect ABXing.  The change is much more subtle in the other 3 files.


Sample S44

Code: [Select]
-------------------------------------
WinABX v0.23 test report
11/19/2005 23:41:52

A file: C:\Documents and Settings\stephen\Desktop\S44\001 S44 [iTunes v6.0.0.18, QuickTime 7.0.3].wav
B file: C:\Documents and Settings\stephen\Desktop\S44\001 S44 [4.2.1.0 VBR internet high].wav

23:42:21    1/1  p=50.0%
23:42:35    2/2  p=25.0%
23:42:46    3/3  p=12.5%
23:43:30    4/4  p= 6.2%
23:43:40    5/5  p= 3.1%
23:43:51    6/6  p= 1.6%
23:44:06    7/7  p= 0.8%
23:44:14    8/8  p= 0.4%
23:44:25    9/9  p= 0.2%
23:44:36  10/10  p< 0.1%
23:45:02  11/11  p< 0.1%
23:45:12  12/12  p< 0.1%
23:45:14  test finished

-------------------------------------
WinABX v0.23 test report
11/19/2005 23:45:15

A file: C:\Documents and Settings\stephen\Desktop\S44\001 S44 [4.2.1.0 VBR internet high].wav
B file: C:\Documents and Settings\stephen\Desktop\S44\001 S44 [3.2.0.15 VBR internet fast].wav

23:45:43    1/1  p=50.0%
23:45:50    2/2  p=25.0%
23:45:57    3/3  p=12.5%
23:46:01    4/4  p= 6.2%
23:46:10    5/5  p= 3.1%
23:46:19    6/6  p= 1.6%
23:46:30    7/7  p= 0.8%
23:46:46    8/8  p= 0.4%
23:46:55    9/9  p= 0.2%
23:47:30  10/10  p< 0.1%
23:47:38  11/11  p< 0.1%
23:47:45  12/12  p< 0.1%
23:47:46  test finished


Some more ABX results for sample S44.  In the "4.2.1.0 VBR internet high", I heard some ugly distortion within the 1st second, which wasn't there in the iTunes AAC or "3.2.0.15 VBR internet fast" samples, hence the perfect ABXing.

Like the first sample I tested, there are some very noticeable issues with the "4.2.1.0 VBR internet high" encoded files that I didn't hear in the other two.  A definite regression on these two samples (S30 and S44).


EDIT:  Added ABX results for sample S44.

Personal evaluation at ~130..135 kbps, 200 samples

Reply #82
QuantumKnot> thanks for testing it (ABX scores is a luxury I didn't request)

If other people could tell what they think, it would be nice. Even if you can't tell any difference, comments would be appreciated. Thank you.

Personal evaluation at ~130..135 kbps, 200 samples

Reply #83
You can choose for yourself what you believe or don't believe on this subject.

Arguing about it seems about as pointless since we can't even agree on what would prove or disprove each others claims.

I will note though, that your current "belief" requires us to have done things such as:

1) Consistently putting the old encoder in "high" mode by default, despite knowing about "fast" mode being overall better without any strings attached.
2) Throw away the most significant improvements from the old encoder and don't use them in the new encoder.
3) Having worked for a long time on the new encoder only to make something that was significantly worse than we had before, even worse than the old "high" mode in quite a few situations.

Personal evaluation at ~130..135 kbps, 200 samples

Reply #84
Quote
You can choose for yourself what you believe or don't believe on this subject.
[a href="index.php?act=findpost&pid=343123"][{POST_SNAPBACK}][/a]

If I'm making listening test it's precisely to not choose what I should believe  I always prefer experimentation to beliefs.

Quote
Arguing about it seems about as pointless since we can't even agree on what would prove or disprove each others claims.

I thought it was precisely the purpose of listening tests... If several people could confirm that the current Nero encoder suffers from specific issue, it would be likely that the issue exists.

Quote
I will note though, that your current "belief" requires us to have done things such as:

1) Consistently putting the old encoder in "high" mode by default, despite knowing about "fast" mode being overall better without any strings attached.
2) Throw away the most significant improvements from the old encoder and don't use them in the new encoder.
3) Having worked for a long time on the new encoder only to make something that was significantly worse than we had before, even worse than the old "high" mode in quite a few situations.


Very sarcastic situation indeed... But is it impossible? If I remember correctly, it was very clear that "in the mind of HA.org administration" (to quote an old recommendation fortunately changed very recently) developers such as Robert, Gabriel, Alexander and Takehiro spent three year of their time to make LAME perform worse than the beloved 3.90.3. Doesn't it mean that developers are not necessary infallible?

Personal evaluation at ~130..135 kbps, 200 samples

Reply #85
Quote
If I'm making listening test it's precisely to not choose what I should believe smile.gif I always prefer experimentation to beliefs.

I thought it was precisely the purpose of listening tests... If several people could confirm that the current Nero encoder suffers from specific issue, it would be likely that the issue exists.


It would be, if the test was over multiple genres with multiple listeners. I am interested in such a test. I am not interest in the sillyness that this one is.

Realistically, you're testing 3 classical samples where the difference in bitrate allocation (despite the known fact that the averages should be the same!) should already be a big warning sign that something is not quite right.

Personal evaluation at ~130..135 kbps, 200 samples

Reply #86
Quote
Second: try to answer correctly please. You perfectly know that the bitrate you get on short samples is not a problem. The previous -internet mode and the current one both lead to similar bitrate (~130 kbps) and if the previous encoder was clever enough to allocate 300 kbps to maintain the quality, then the current one has issue.


It might be suboptimal on these samples. So what? How does this allow you to make any kind of conclusion by letting someone else confirm that a sample encoded with 3x the bits is better?

Personal evaluation at ~130..135 kbps, 200 samples

Reply #87
i just tested s22 and was only able to abx it with 8/10 against nero 4.2.1.0. the other samples were transparent to me.

my ears are not well trained, i just try to hear things. i don't want to concentrate and relisten to a sample 100-times, just to pass a test. i think it get's too boring. i won't hear specific issues while i just listen to the music (mostly in my car) anyway
member of the "i have a cat-avatar"-group ;)

Personal evaluation at ~130..135 kbps, 200 samples

Reply #88
Quote
It would be, if the test was over multiple genres with multiple listeners. I am interested in such a test. I am not interest in the sillyness that this one is.[a href="index.php?act=findpost&pid=343129"][{POST_SNAPBACK}][/a]

I know. You repeated several times in the last days that such tests are not "proper". Your opinion, not mine.

Quote
Realistically, you're testing 3 classical samples where the difference in bitrate allocation (despite the known fact that the averages should be the same!) should already be a big warning sign that something is not quite right.

You probably missed the 197 other samples. I can't upload a full set of 200 samples encoded with two or three encoders and asking to other members to test them. That would be silly. And as I said, these three samples are just here to illustrate one problem I discovered. I never said that Nero Digital has regressed everywhere (it would be silly), but just that some enhancement that were appreciable with the previous 'fast' encoder are not reported to the new LC encoder which was yet supposed to be this old fast encoder.

Personal evaluation at ~130..135 kbps, 200 samples

Reply #89
Quote
It might be suboptimal on these samples. So what?
[a href="index.php?act=findpost&pid=343131"][{POST_SNAPBACK}][/a]

  Is this question serious?
BTW, it's not "it might" but "it is".

Quote
How does this allow you to make any kind of conclusion by letting someone else confirm that a sample encoded with 3x the bits is better?

You made a claim (that most people won't perceive a regression), and is it to me to proove that you're wrong? As I said, I only post three samples as example, just to see if the regression I heard are something fictive and/or personal, or if these regressions are also audible to other people.
What I'm trying to prove is that Nero Digital v.4.2.1.0 has problem that Nero Digital v.3.2.0.15 didn't have with 'fast mode'. It's perfectly true that the previous encoders bloats the bitrate, but at least this VBR mode iwas doing what VBR are for: keeping a constant quality. I can't say the same for current VBR -internet mode. And two persons confirm it.
BTW, the S44 sample correspond to the "biniou sample" I submitted long time ago. The problem was corrected once, but is now resurrected with the new encoder.

Personal evaluation at ~130..135 kbps, 200 samples

Reply #90
Quote
You made a claim (that most people won't perceive a regression), and is it to me to proove that you're wrong?
[a href="index.php?act=findpost&pid=343140"][{POST_SNAPBACK}][/a]


I remember quite well what happened the last time I did want to prove my claim (32kbps HE-AAC): I wasn't allowed to.

Quote
As I said, I only post three samples as example, just to see if the regression I heard are something fictive and/or personal, or if these regressions are also audible to other people.


You are still missing the point entirely: I don't disagree specific samples can be worse. I disagree the encoder is overall performing less good, that is, on MULTIPLE GENRES and MULTPLE LISTENERS.

Since you now even seem to disagree that is an important distinction ("You repeated several times in the last days that such tests are not "proper". Your opinion, not mine."), I have nothing more to say.

Personal evaluation at ~130..135 kbps, 200 samples

Reply #91
Quote
Quote
You made a claim (that most people won't perceive a regression), and is it to me to proove that you're wrong?
[a href="index.php?act=findpost&pid=343140"][{POST_SNAPBACK}][/a]


I remember quite well what happened the last time I did want to prove my claim (32kbps HE-AAC): I wasn't allowed to.
[a href="index.php?act=findpost&pid=343145"][{POST_SNAPBACK}][/a]

Come on. You're allowed to do it (didn't you performed it in fact?  ), and you know it. You just can't set a test with wrong contenders and choose you own samples.

Quote
You are still missing the point entirely: I don't disagree specific samples can be worse. I disagree the encoder is overall performing less good, that is, on MULTIPLE GENRES and MULTPLE LISTENERS.

Is it with me that you disagree on this point? Did I said that the new Nero is generally worse than the previous one?  If the answer is positive, could you give me the corresponding link? Thanks.

Personal evaluation at ~130..135 kbps, 200 samples

Reply #92
Garf, Nero AAC may to some extend be your baby.... however, a certain kind of reaction from devs towards listening tests (i'm primarily not talking about the "what" but the "how") torpedoed musepacks public image on ha.org not so long ago.

I asume i'm not alone with this impression when i say that AAC has been in the making for a LONG time, it has been sold to us as the "next big thing" and "the future of lossy encoding".... and when during the previous years the tests didn't look good, we were told over and over "well, its still new and requires some more tuning and more research, then it will rock"..... with nero AAC, this is what we've been told for almost EVERY upcoming version: "yes, there are some errors, but the next version will be much better and probably fix this and that"....

Now, when after years in the medium bitrate arena quicktime AAC can barely compete with vorbis, which does NOT have many of the patented toys available to AAC...... and nero AAC can barely beat LAME-MP3.... the format which supposedly is "obsolete"...... then maybe it's no surprise that people aren't very euphorous about the performance of AAC, especially nero AAC? Or is this encoder only gonna be useful for narrowband-scenarios? (question is intentionally worded in a provocative way)

- Lyx
I am arrogant and I can afford it because I deliver.

Personal evaluation at ~130..135 kbps, 200 samples

Reply #93
guruboolez great test, you trully have the patience of Job.

Also, Garf claimed several times that "most people" would perceive new encoder as better, is he referring to the old (Roberto and Co.) listening tests or is there another study I've missed? Anyone has a link to the study in which those "most" people participated?
The Plan Within Plans

Personal evaluation at ~130..135 kbps, 200 samples

Reply #94
Quote
guruboolez great test, you trully have the patience of Job.

Also, Garf claimed several times that "most people" would perceive new encoder as better, is he referring to the old (Roberto and Co.) listening tests or is there another study I've missed? Anyone has a link to the study in which those "most" people participated?
[{POST_SNAPBACK}][/a]

Nobody has so far participated in a listening test proving that "most people" are considering the new encoder as better than the previous one. Not even me. Not on HA.org, Doom9 or any other known board or website.
He's only referring to his own beliefs, disguised as valid and general claims.

BTW, latest listening collective test involving Nero Digital LC-AAC was performed with [a href="http://www.rjamorim.com/test/aac128v2/presentation.html]Nero AAC 2.6.2.0[/url]. It was 21 months ago. I don't think that Garf's reference was this very old encoder when he talked about progress of the latest one. He probably refered to the last one of the 3.2.XXX series.

Personal evaluation at ~130..135 kbps, 200 samples

Reply #95
Guys... I really think this discussion is not getting anywhere

I suggest to wait 10-or-so days for the multiformat ~128 kbps listening test, where we all should take part - and also Nero AAC will be upgraded with some necessary quality fixes collected during the testing.

With sufficient number of trained listeners, and multiple genres - we'll be able to tell the actual state of audio compression and how much codecs did improve during recent time.

Personal evaluation at ~130..135 kbps, 200 samples

Reply #96
If you have some time, check the current highest VBR mode "transcoding", which often provide a lower bitrate than 2nd best preset ("audiophile"). Low volume parts are apparently affected; and it may have an audible impact when (a very strong) replaygain track mode is applied. See for example:

http://guruboolez.free.fr/ENSEMBLE/E20_MOD...rings_quiet.ofr
http://guruboolez.free.fr/VOICES/V03_CHORUS_Female_A.ofr

With V03, even --extreme preset (two steps below transcoding!) has higher bitrate (and quality also). I suppose that's a bug or a tuning issue, because currently Nero Digital "transcoding" sound worse than LAME 3.97b1 -b128 with this sample²

² after replaygaining.


EDIT: changed --normal to --extreme

Personal evaluation at ~130..135 kbps, 200 samples

Reply #97
Well, next week is gonna be busy

Personal evaluation at ~130..135 kbps, 200 samples

Reply #98
Thank you very much for going through the FSB 3rd degree questioning ordeal (FSB is the successor to KGB, btw)

Your comments about methodology and motivation were very helpful, thank you again.

BTW, by detection accuracy I was merely referring to absolute tresholds of detection.

I've noticed myself, that just upgrading headphones made previously completely unaudible distortions audible to me.

Hence, at least on certain type of distortions, the treshold was clearly lowered, enabling me to hear and grade artifacts that would have gone unnoticed previously.

regards,
halcyon

Personal evaluation at ~130..135 kbps, 200 samples

Reply #99
Quote
I've noticed myself, that just upgrading headphones made previously completely unaudible distortions audible to me.
[a href="index.php?act=findpost&pid=343192"][{POST_SNAPBACK}][/a]

If I replace my Beyerdynamic with a set of cheap erbuds (Sennheiser MX550), several distortion cease to be audible (but not pre-echo, still perceptible). But I'm not fully convinced that a very expensive of headphone (like Stax ones) would really really help me to catch additionnal encoding issues.