I've uploaded 4 samples:
Fatboy encoded with 3 encoders @ 192 stereo psy model 1.
You'll find the original wave file and 3 codecs : SoloH, MP2 codec found in CDex (version1.13) and Toolame...

here are the samples:
MP2 Challenge !



PS: later I'll upload castanets I think...

Thank you for advance for your cooperation !

Not sure than fatboy.wav is the best sample for an evaluation.

sample 1 = no immediate problem
sample 2 = chocking in comparison : very strog lowpass, and smearing too
sample 3 = a bit flat compared to sample [1], and not as crisp. But nothing annoying
sample 4 = same thing as sample 2, but additionnal chirping => worst of the four

1... 5/5 (the original, I suppose ?)
2... 2.5/5
3... 4.5/5
4... 1.5/5

what's the best IYHO ?

Try with something like 41_30.wav
Cymbals are more common in pop/rock/jazz music

Castanets Samples uploaded :A,B,C and D.
1,2,3 and 4 are still fatboy.

CASTANETS.WAV :

Sample A = nothing disturbing
Sample B = strong lowpass. Guitar is distorted. Pre-echo is annoying. Annoying artifacts (pop) at second 1:0 on left channel
Sample C = smearing. Guitar is flat, and attacks are not sharp
Sample D = less flat as [3]. Guitar a bit distorted, as [2]. Smeared attacks and annoying artifacts (pop) at second 1:0 on left channel

A... 5/5
B... 2.5/5
C... 2.9/5²
D... 2.5/5

² Can't really explain why, but I find this encoding better than the two others. Less rich, but maybe more clean.

Death2 samples uploaded
MP2 Challenge

Fatboy:
1. 10/10
2. 8/10 (less depth)
3. 8/10 ( -"- )
4. 4/10 (very noticeable ringing)

Castanets:
1. 10/10
2. 7/10 (less depth [esp. guitar])
3. 6/10 (slightly muffled)
4. 5/10 (pop + all of above artifacts but slightly worse)

Death2: (what's this kind of music called?)
1. 10/10
2. 8/10 (slight ringing at clicks+slightly distorted noise at the begining)
3. 2.5/10 (sl. r.+distorted noise at the end+something weird in the background - more noticeable than 4)
4. 3/10 (sl. r+muffled n.+sth. w.)

Well, could you post the coders on your site, or is it illegal?
I'd like to encode some of my testing samples too...

thanx to AstralStorm & guruboolez for their cooperation !

I need more participants please !!
(fatboy/CDex/Toolame).

please,
can anyone ABX those samples ?
thx for advance

I need statistics...

sincerely,
nick

I cannot hear the difference, my ears are not very good, just sample 4.flac (fatboy) has very bad artifacts and sounds much worse.

Have you also tried InterVideo Audio Encoder? They offer an DirectShow-Filter encoding MPEG Layer I, II and III, AC3 and AAC. I just found it. Unfortunately everything except MPEG Layer I / II is locked in the filter. The settings differ from the normal iso-based encoders, you have 4 quality/speed settings, dual/stereo/joint/mono setting and the bitrate. No psy-model switch.

For AAC you have Intensity, Mid/Side, PRED, TNS, Short Window, Main/LC/LTP/SSR, LFE (yes, it allows multichannel AAC), MPEG2/MPEG4 and Bitrate setting.

Unfortunately, Rule Nr.9, but maybe you want to search for a FourCC of a videocodec on http://www.fourcc.org/fcccodec.htm ? You didn´t found the FourCC?, maybe you´ll find it in the Ukraine.

Is it really legal ?!

I don't think so !
sincerely,
Nick

Maybe in the Ukraine!

I encoded both samples with InterVideo, 192Kbit/s, High Quality for you if you don´t bought ( ) the the DS-Filter.
Maybe you can mix them into the others.
iv_fatboy.mp2 (116KB)
iv_cast.mp2 (154KB)

thank you for providing me those samples S_O !!



the first: sound different from the original, preecho...
I prefere CDex or SoloH...
castanets: not enough time time to hear it...


Sincerely,
Nick

PS: I'll be back friday...so I try to test this encoder (aka try winproducer trial).
PS 2:One more time: thank you !

Just FYI, SoloH is pure dist10 code compiled into a fancy GUI.

I would also suggest using some other MP2 encoders:

MAenc - Seems to be another derivation of dist10

SCMPX - another dist10 suspect. Like MAenc, I never bothered to check if it's really dist10 or if there are differences.

Both of the above encoders are free.

Qdesign - commercial, quite famous. Works as a Windows ACM module.

If you want me to provide samples using QDesign, just say so.

Regards;

Roberto.

Based on the sample-space of the replies (2 people?), what was the preferred option?

enquiring minds want to know.
mike

Original sample is 1 and A.
CDex is 2 and C.
SoloH is 3 or D.
Toolame is 4 or B.

AstralStorm:
Fatboy:
1. 10/10 WAV
2. 8/10 (less depth) CDEX
3. 8/10 ( -"- ) SoloH
4. 4/10 (very noticeable ringing) Toolame

Castanets:
1. 10/10 WAV
2. 7/10 (less depth [esp. guitar]) Toolame
3. 6/10 (slightly muffled) CDex
4. 5/10 (pop + all of above artifacts but slightly worse) SoloH


IMHO I would say:
1- SoloH
2- CDex
3-Toolame

May I ask why using of psy. model 1?

Psy 1 because you get higher hi-freq response and IMHO is better in tems of quality compared to Psy 2.
Yes, 192 kbps is a bit short, 224 or better would be recommandale but it's easier to find artefacts @192.

Psymodel 2 is usually considered much better, quality-wise, than model 1.

That's the reason, for instance, that MPC is based on model 2

Unfortunately, tooLame limits frequency of model 2 to 17kHz. It's a bug in tooLame, not a limitation in the model.

Just one thought - may be wrong!

I had Eric Claptons Unplugged DVD here.
The AC3 stream at 448kbit sounded that artifacted sometimes
it even didn`t do much fun to listen. The LPCM stream decoded to mp3
was much superior.

Now i am totaly new to DVDs AC3 and this LPCMs and so i may be wrong!?
Isn´t AC3 mp2 in general?

Wombat

No. AC3 quality is considered similar to MP2, but they are very different coding standards.

Thank rj to tell me i´m wrong.

When all this was so easy, there would be no fun left.

Wombat

rjamorim >
well, Psy 1 sounds better to me (SoloH and old CEP MP1/2 filters).
Psy 2 is a bit dull, aka, cuts @16kHz.

May be I'm wrong
and some guys here said Ps2 is better medium BR and Psy for higher BR,
that's why I would choose Psy1.

Sincerely,
Nick

Certainly in toolame, psy model 1 is much better than 2 around 128kbps. Also around 192kbps stereo is much better than joint stereo. Maybe even at lower bitrates for some samples. I haven't found any audible difference between psy1 and 2 at 192kbps using stereo in the tests I've done.


IIRC Psy 2 was broken in the old CEP mp2 filters and produced really very bad results.


At 128kbps I've found SoloH can be surprisingly good on some test samples, but terrible with typical music.


I'll download your samples and have a listen, but I probably won't be able to hear the difference. I was put off downloading because it appears that you decoded the mp2s to .wav and then re-packed them using FLAC - why not just make mp2s available and save everyone some download time? To hide the original? We could do that ourselves with ABC/HR.

Cheers,
David.

EDIT: Now I've downloaded the files, I realise the filesize wasn't such a big issue. Still, some people still use modems you know!

I used ff123's ABC/HR. No ABX needed. I knew that Samples 1 and A were the originals, so I loaded them as such.

Fatboy:
Sample 2: CDex: 4.0
Sample 3: SoloH: 4.0
Sample 4: TooLame: 2.5 - this one has much more tinkling sounds
Sample iv: InterVid: 4.0

Castanets:
Sample B: TooLame: 2.5 - There's a loud "blip" in the left hand channel during the first few castanets
Sample C: CDex: 4.0
Sample D: SoloH: 3.0 - There's a loudish "blip" in the left hand channel during the first few castanets
Sample iv: InterVid: 4.0


I can't differentiate the 4.0s in either tests - roughly equal amounts of pre-echo, no other obvious problems to me.


I was surprised by the results. I'd already tested toolame with fatboy, and it didn't sound that bad. So I went back and re-checked.

I suspect that you used psy model 1 or 2 with joint stereo at 192kbps to encode at least the toolame fatboy sample, giving the tinkling or metallic effect that I heard. You reported that you use stereo, but using stereo, there are no loud "tinkling" or "metallic" problems. There are other problems, but these are less obvious to me.


I wonder what settings you actually used for the other encoders? If you have compared stereo with joint stereo, then it's hardly a fair test!

Cheers,
David.

david,
thnk you very much for testing the samples !

1/ why not just make mp2s available and save everyone some download time? To hide the original?
you're right !

2/ I suspect that you used psy model 1 or 2 with joint stereo at 192kbps to encode at least the toolame fatboy sample
well ! to my mind, all tests were done with joint stereo, but I must admit I could have done a mistake , but if you want to be sure, you can download the fatboy sample and encode it @ 192 kbps, JS and Psy 1.
Sorry for the inconvenience if I've made a mistake (original MP2 files were deleted, I'm stupid ).

3/ I wonder what settings you actually used for the other encoders?
same settings, aka: 192 JS Psy 1.


Thank you very much for your cooperation !
Sincerely,
Nick

Damn !

this is not JS but S !!

ALL TESTS WERE DONE IN STEREO !
sorry

No, Nick, you originally said that all your tests were done in stereo (first post), but the fatboy toolame sample is definately joint stereo.

Cheers,
David.

P.S. It's silly to use lossless packing when doing a comparison of same format lossy codecs - just trust the testers, and give them a lossless original, and the lossy version without decoding. You're not hiding which is the original anyway - 1 second in spectral view in any audio editor will make it obvious! So if people want to cheat, they will cheat. And if they don't, they will do blind testing themselves. Just my opinion - others may not agree.

Hum I was sure I've used Stereo !!
But as you say, Fatboy is joint stereo, I've made a mistake...
Sorry...

I'll correct this sample as soon as possible !!!

Well,
the page is now corrected,
thanks to David Robison's help !!

Now you can get the results in results.txt and Fatboy samples encoded w/ 2lame is now REALLY stereo:
-p 1 -b 192 -m s double checked !

Sincerely,
Nick

All should now be stereo !

i mean David Robinson and not Robison... B)
Sorry David

Hi all,
I know listening tests are the 'ultimate' diagnosis on the quality of a psychoacoustic encoder, but take the time to have a look at the source of these encoders. When I find some time (probably never) I'll sift through them all.

CDEX's MP2 encoder for example is a branch off toolame from the early 02 days. I don't think any quality issues have been addressed in the PAMs. For starters, one them is mine, with my comment on it:

cdex/psycho_0.c
/* a bunch of SNR values I sort of made up MFC 1 oct 99 */
static FLOAT snrdef[2][32] =


AFAIK SoloH (which is as old as the hills) is just a straight compile of the dist10 source. No magic "uber-PAM" in there.

While toolame certainly hasn't improved markedly on any of the psy models, PAM3 and PAM4 in toolame corrected bugs within the code.

In reality, any dist10 based mp2 encoder should probably sound almost the same as any other. Any differences in sounds are probably due to bugs/compiler errors rather than any difference in their PAMs.

just my thoughs
mike

Nick Jr III, why don't you use psymodel 2 on toolame, if this psymodel works better at these bitrates? A cut-off at 17 KHz is not audible in 95% music or more.

Nick Jr III, why don't you use psymodel 2 on toolame, if this psymodel works better at these bitrates? A cut-off at 17 KHz is not audible in 95% music or more.

Well KikeG,
I sincerely think PAM1 is better than PAM2 due to the lower cut-off of this one...
and this cuff-of is about at 16/16.5 kHz, but I must admit PAM2's testing could be interresting !

May be I'll use PAM2 later to try !
Thank you for your remarks !!

Nick,