As promised, the audio samples I used for my own listening test (http://www.hydrogenaudio.org/forums/index.php?showtopic=14091&hl=).
It would be nice if others people will test themselves, in order to compare results.
Note :
• Due to my limited bandwith, I can't upload the whole gallery in one night. I'll try to find a DSL connection this week. Be patient.
• For the same reason, I didn't choose flac as lossless format, but OptimFrog, more efficient (31.6 MB against 36.2 MB). OptimFROG is not open-source, but is available on linux plateform. Playback & decoding is easy on Windows, with foobar2000 & Winamp plug-in.
Vocal and orchestral sample, from an Haendel Oratorio.
Cello samples (cello and light continuo), from Antonio Vivaldi
String quartet on period instruments, from a compser called Gossec.
Orchestral : Music for Royal Fireworks (G.F. Haendel)
Instrumental (violin solo and accompagnato), from Four Season of Antonio Vivaldi :
Mozart, orchestral ouverture :
-- removed for the moment --
Serail.ofr is now clean and complete
Rest will follow this week, or during the next week-end.
Dorilla sample is problematic with Nero AAC encoder ; very specific artifacts. If someone want to confirm...
Another one, for Ivan:
(Giulio Cesare, from Haendel, conducted by Marc Minkowski)
(obvious problems with latest experimental Nero encoder)
(obvious problems with latest experimental Nero encoder)
Latest is 2.5.7.6. Did you test with it?
I don't notice any noticeable problems. The original sounds weird too especially around 3s.. Could you be more specific?
Attached is the 2.5.7.6 version
Yes, with -streaming profile, and on two computers tested, result was absolutely awful (for my taste). I've just sent an email to Ivan, but it would be nice if you can test it too.
Hornpipe is very bad too (same kind of distortions).
Less annoying, Inverno is strangly smeard for a so high bitrate (near 160 kbps IIRC).
Yes, with -streaming profile, and on two computers tested, result was absolutely awful (for my taste).
What is the problem exactly? Imo it's not so awful.
I upload 2.5.7.6 version of Questro above, check that out to be sure we are testing the same version...
Artifacts are the same. Can't explain : I would call it flanging, on strings behind trumpets. With Questo_Core, adding tiny harpsichord in continuo, is more distorted too. Did you try? What do you think?
Err.. what trumpets? I can't hear any in Questro - or is something wrong with me
Well I can't hear anything wrong with these samples
So, this leaves us few options - Guruboolez has a bunch of daily builds and he can (if he has time of course) tell me if this was introduced all of a suden in some version - could be a bug - but neither my ears, nor PEAQ can mark these samples as problematic... I am really puzzled
Artifacts are the same. Can't explain : I would call it flanging, on strings behind trumpets.
Do you mean with hornpipe.
Please explain when you upload a file: what you hear wrong, and what time section.
Although I'm not a strong believer of odg, the hornpipe odg is good: -0.46
I do know that this odg doesn't mean that it couldn't be awful, but I wouldn't rate any of these "awful".
I spoke about trumpets, in order to locate the artifacts on the sample JohnV uploaded: Hornpipe.
On questo_core, there are no wind instruments : mainly strings (and vocal). Artifacts, for me, are distorting strings. Sound is hollow for me: really bad.
I played it on foobar2000 0.72 alpha 8 - I will check it on QT or iTunes.
I'm just talking abour -streaming profile, HIGH quality.
I will test it with others aacenc32.dll versions. Just be patient
Guruboolez,
Can you upload decoded files somewhere? This is really interesitng.
Here are both the original and decoded hornpipe.
Can anybody verify the very bad artifacting of trumpets/horn there (I assume the 20-24?
I'm worried, because I do not hear anything very bad there.. I didn't ABX though, but since I don't hear anything very wrong, I didn't even try yet (I just can't hear anything "very bad"). Also odg shows very good figure and the trumpet section has very very small difference when doing inverse mixing with the original.
Either Guru has truely very exceptional hearing or there's something wrong somewhere...
I've just finished my tests. I will see later for the decoded file (please, precise for which encoder).
##1 ##2 ##3 ##4 | slow
|
2.5.5.8 2.5 2.5 2.0 2.3 | 2.5
2.5.6.2 2.5 4.0 3.0 3.3 | 3.5
2.5.6.3 3.0 4.0 2.5 2.8 | 3.5
2.5.6.8 2.0 1.5 2.0 1.0 | 1.3
2.5.6.9 2.9 1.0 1.0 1.0 | 1.0
2.5.7.4 2.5 2.0 1.0 1.0 | 1.4
2.5.7.5 1.0 1.0 1.6 1.0 | 1.4
2.5.7.6 1.1 2.0 1.6 1.5 | 1.3
The four first tests were made very quickly: I called them blitztest.
Purpose of these tests are to see if I'm able to immediatly detect the encoded one, and to give a very approximate notation.
Hierarchy isn't reliable, because perception chage a lot between first file and last one.
Nevertheless, I tried to be slightly more precise on test #3 and test #4.
It appears that two encoder are really better : 2.5.6.2 and 2.5.6.3. Both are followed by official 2.5.5.8
All encoders released with and after 2.5.6.8 are in my mind horribly distorted.
With the lest test I performed the notation was more precise. I took my time for listen them many times, in order to draw a precise hierarchy.
Again, two winners I couldn't differenciate (2.5.6.2/6.3) and another one, behind but less annoying than all others (2.5.5.8)
The five last releases I have are clearly worse here. 2.5.6.9 is possibly worse than others. I'm not really able to differenciate four of them.
P.S. I focused my attention on two first seconds (blitz test) of QUESTO_CORE sample, the more destroyed. Some parts of this sample are not "awful" with latest encoder. Nevertheless, my overall impression is very, very negative.
It is really confusing to me now -
Differences between 2.5.6.9 and 2.5.7.6 (both were release candidates for next Nero) are only in some bug fixes that can't be related to VBR mode in any way. I'd certainly do a double check but it is really starting to sound -- spooky
I forgot to mention this : all files, including reference, were replaygained, and decoded with dithering enabled.
Here is the encoded one, replaygained/dithered with latest FB2K, used for my test:
Right, with questo_core I hear some problem just before 3s and can ABX, but imo if you give based on this score 1.1, it is in no porportion to anything more serious problems. I think you need to use larger scale...
Or maybe you just are so familiar with this music, I don't know...
It is really confusing to me now -
Differences between 2.5.6.9 and 2.5.7.6 (both were release candidates for next Nero) are only in some bug fixes that can't be related to VBR mode in any way. I'd certainly do a double check but it is really starting to sound -- spooky
Even if the tests I made were performed very quickly, and forbid strong conclusions, I would say that 2.5.6.9 and 2.5.7.x are in the same class.
2.5.6.2 and 2.5.6.3 are different, and better (at least, here).
Something happened between .6.3 and .6.8.
I must test 2.5.7.7. now
Hello!
I was just curious, didn't know much about my hearing abilities.
I've ABX'ed the hornpipe files just a few minutes ago. i did only a short test.
But a lay my hand on the holy bible and swear I've not cheated.
I've scored a 5/5 on the test. It was quite hard and I stoped ABXing due to lack of concentration.
I recognized more distortion on the whole track, sounding to me like background noise.
Hardware I used: Soundblaster AWE64 Gold, Sennheiser HD-590
Here are both the original and decoded hornpipe.
Can anybody verify the very bad artifacting of trumpets/horn there
I've tried to ABX Hornpipe.ofr (guruboolez' original) vs. Hornpipe.mp4 (Hornpipe.zip) - I hope it's the same as "hornpipe comparison.zip.
Focussing on 17.4-19.1 I was able to ABX 8/8, the attacks of the three fast trumpet tones sound noisier, airier, less sharp; probably pre-echo. Its not noticable for me without direct A/B comparison of short sub-samples. I guess this is not the problem that needs to be verified.
Maybe I've notices something that sounds more like what guruboolez has described at 20.2-23.4 but I haven't been able to ABX successfully yet.
what are these ##1 ##2 ##3 ##4 and slow? Are these different positions or different rounds or what? And are you using the same sample all the time?
For example with 2.5.6.9 ##1 is 2.9 but ##2 is 1.0 ? It goes from "slightly annoying" (3) to "very annoying" (1)??
naturfreak: The issue here was not if you can ABX the file overall or not. I don't think it's really hard and not the point here. The issue here is do you hear very bad distortion during 20-24s with the trumpets/horn with hornpipe?
naturfreak: The issue here was not if you can ABX the file overall or not. I don't think it's really hard. The issue here is do you hear very bad distortion during 20-24s with the trumpets/horn with hornpipe?
Sorry, JohnV.
I will check it.
Yes there are more distortions in one of the files if I can rely on my ears but I won't describe them as "very bad". (Section 20-24 s)
Oups, sorry for being so confuse.
##1, ##2... were the four bitztests I made in a first time. "slow" is the fifth and last one, with a more care in notation (hierarchy). First blitztest (##1) was the fastest one : notation isn't reliable at all. No need to compare or to take in count the the notation of the two first tests.
JohnV: if you have the 2.5.7.7. could you compare the extreme beginning of englese.wav. I ABX with successfully 2.5.7.6 and .7.7: kind of grainy sound, something dirty...
The following samples are for Ivan Dimkovic. They're all under 30 seconds, and were extracted from classical discs. I've no other place for upload them.
Various artifacts appear with aacenc32, with streaming profile (2.5.8.2. & 2.5.8.4.). Lame ABR [134] is sometimes better (Zarzuela.wav - Eryximachus.wav - Hérodiade.wav).
Files are tagged, and the COMMENT field contains the ABC/HR log file of ABX tests & comparison between nero AAC, iTunes AAC and lame MP3.
Results obtained with these samples are not representative: I've selected these 7 samples on problems I heard with Nero encodings.
(please, do not comment my notation: there's no relation between the level of the degradation and the level of my irritation)
Four short samples (Orchestral - String quartet - Gamba)
Lyric one, with problems on trumpets.
Complete sample: could integrate a listening test at < 150 kbps.
Harpsichord with flute and gamba
Last one, similar to the previous file :
Different from the another, here is an orchestral sample I've ABXed with :
- iTunes [QT 6.4] AAC at 224 kbps
- Nero AAC [2.5.8.4] ::audiophile::
- MPC [1.14] --standard and --insane
- DualStream [4.507] --quality 3
[short details are in COMMENT field of the tags].
The little problem with mpc annoys me, because the incriminated passage doesn't seem to be really hard to encode. I can't say that artifacts are annoying, at least with mpc insane, but at --standard the difference was easy to perceive, without direct comparison.
P.S. In the sample called American (in the Four Samples archive), there's a noise amplification with Vorbis 1.01 I've ABXed at -b 8. It's very obvious at -b 4, obvious at -b 5, and less obvious but still audible above.
Related to the patched version of ogg vorbis, the cello sample mentioned in this thread:
http://www.hydrogenaudio.org/forums/index....=0&#entry167658 (http://www.hydrogenaudio.org/forums/index.php?showtopic=16903&st=0&#entry167658)
Artifact (boiling sound for me is on seconds 2-4). Noise is lowered too with the patched version.
for Roberto: 19 lyrical samples encoded at 96 kbps (he-aac). For the next multiformat test.
Four classical samples for the multiformat test. (MPC --standard)