MP2 / Resampling Listening test

[KpeX]

Hi SVCD users,

A recent thread in the audio encoding forum brought up some interesting issues on the quality of audio and what can be discerned. This is the first of hopefully a series of test on what can be discerned in movie soundtracks under real-world encoding conditions. The first test is simply to test whether users can discern artifacts from the resampling of audio.

There are two files in the zip file linked below, arbitrarily named; both are a 1 minute sample of a 128kb/s CBR mp2 encoding transcoded from 5.1 AC3 using BeSweet with azid and 2lame. One of the files was transcoded directly to mp2 keeping the 48Khz sampling rate throughout the process. The other file was resampled twice before mp2 encoding, first from 48 khz to 44.1 khz and then upsampled back to 48khz. This in theory should have introduced twice the rounding error from the standard 48 khz > 44.1 khz resampling.

Please reply with your ability to discern between the files, and which one ( file1.mp2 or file2.mp2 ) has gone through the resampling cycle, not in this thread, but in email to kpex_03@yahoo.com . Even if you can't discern between the two, please reply because the data will help to get a statistical average. Please only reply to this thread with test questions or comments, not your guesses or predictions as to which file is which. Please only download if you truly intend to test, my bandwidth is limited.

http://www.freewebs.com/kpex/MP2_Challenge.zip [Right click and save target as - DDogg edit]

Mirror 2: http://varied.myby.co.uk/shared/MP2_Challenge.zip

Regards,

[jorel]

sorry to post here KpeX
but after 24 hours,i can't download the file(more than 20 tents).
(i see your post a few minutes after you wrote)
i try also like DDogg recomendations but don't work!

2 favors please:
another link to download and
delete my (this)post !

thank you!
:)

[KpeX]

That's strange. Download link is still working fine for me. PM me your email address and I'll get it to you. This goes for anyone else that can't download as well.

[r6d2]

Originally posted by jorel
i try also like DDogg recomendations but don't work!

I had the same problem. Fixed by disabling "Click monitoring" option in GetRight.

I usually have to do that when saving files from this forum. Don't know why. It works for small downloads though.

@KpeX, how is it doing? Got several answers?

[KpeX]

3 responses so far, definitely not any significant statistical information, so please do keep the testing coming for anyone who hasn't tried yet and is interested :).

[KpeX]

New mirror up here: http://varied.myby.co.uk/shared/MP2_Challenge.zip

Thanks to Imperial Llama for the webspace.

[r6d2]

Originally posted by KpeX
3 responses so far, definitely not any significant statistical information, so please do keep the testing coming for anyone who hasn't tried yet and is interested :).
I hope you at least got jorel's surround ears reported :D

Come on guys! :)

If you're a visitor of this forum it's because you like challenges, don't you?

Or you thought that DVD backups were a piece of cake? :D

[DDogg]

KpeX, here (http://forum.doom9.org/showthread.php?s=&threadid=17001&highlight=joint+stereo) is what I was trying to remember about how stereo has to be used with mp2 dolby downmixes.

Quote from "Cokes"
"But what DVD,SVCD a VCD concerns (MPEG-1(2) Layer 2 audio): Layer 2 doesn't know anything about MS, it uses only IS, so if you encode your Dolby Surround donwmixed audio in Joint Stereo mode of any Layer 2 encoder - it destroys surround information."

[KpeX]

Yep, that confirms some threads on hydrogenaudio.org found by myself and Boulder (discussed in the audio encoding forum) that confirm, basically, that MP2 joint stereo is not very advisable. Very interesting stuff.

[jorel]

Originally posted by r6d2
I hope you at least got jorel's surround ears reported :D


sorry for late,i have internet problems here!

@ KpeX
thanks,the second link works fine,i got it now!
the audible diferences from 48k and 44k(re encoded to 48k)
i will mail to you in minuts!
:)

@ r6d2 and KpeX

first impression:
file1 have more volume than file2 in right channel.
file2 have more volume than file1 in left channel.

left channel in the 2 files have some "clips" (distorcions),
the general quality is reasonable but deserve more diligence!
...like i wrote:
128(like in samples)is only for conversations,192 for musicals!

i'm searching words in dictionary to explain,
the idea is to be honest,not offensive:
:o
for me 48k is better comparing with 44k,
seems "no sense" re encode the audio to 48k,
again to 44k....and again to 48k !
:eek:
no matter what sounds better in this case,
i never did it cos is useless.

what i ever do is compare 44k with 48k
---> 48k is better(with 192 is really better)!
all my encodes are using 48k!
:)

edit:
i forgot:
was used 2lame like you posted.
2lame it's better for 44k,not for 48k!
;)

try the same with Headac3he,downmix surround2,
2 pass mode,dual channel or stereo!

[jorel]

Originally posted by KpeX
Yep, that confirms some threads on hydrogenaudio.org found by myself and Boulder (discussed in the audio encoding forum) that confirm, basically, that MP2 joint stereo is not very advisable. Very interesting stuff.

i post about it too,please read my post here:
http://forum.doom9.org/showthread.php?s=&threadid=64340&perpage=20&pagenumber=1

"stereo and/or dual channel are better than joint stereo
(for prologic receivers and ..."

and i post this link too:
http://andreas.welcomes-you.com/projects/dv/index.html#jointstereo

!

[r6d2]

Originally posted by jorel
for me 48k is better comparing with 44k,
seems "no sense" re encode the audio to 48k,
again to 44k....and again to 48k !
Of course, my friend, but the idea was to see if the distortion was audible. If KpeX had left one of the files at 44 you'd have guessed by picking at the file and not by listening (cheating :))
:o

[KpeX]

for me 48k is better comparing with 44k,
seems "no sense" re encode the audio to 48k,
again to 44k....and again to 48k !
r6d2 is correct. Additionally, the theory is that resampling introduces quality loss by rounding errors when recalculating sample points. Therefore resampling to 44.1 and then back to 48 will introduce twice the quality loss as resampling once. So if you can't discern these two files there's no way you can discern 44.1 resampled vs. 48 khz.

And as far as physically being able to hear 44.1 orginal material vs. 48, there is really no realistic ability to hear the difference between these two. The reason 44.1 khz was chosen for CD is because it is high enough of a sampling rate that virtually noone can discern between that and higher sampling rates.

[jorel]

KpeX wrote:
"Therefore resampling to 44.1 and then back to 48 will introduce twice the quality loss as resampling once...."
depend what was used to encode.
take a read on audio faq from Doom9
(that was edited for you a few days after this test):
15) Is it better to downsample my DVD audio to 44.1 KHz or keep it at 48 KHz?
http://forum.doom9.org/showthread.php?s=&threadid=7633

" The reason 44.1 khz was chosen for CD is because it is high enough of a sampling ....."

this is not true,you need to read more
(much more)why 44.1 was choosed.
eletronic was my work for 30 years
and don't confuse my horrible english with my
knowledge with eletronic,image and sound!

well...the strange of it all is:
first you post that don't believe me:
" I honestly doubt you can hear the difference that clearly between 48 and 44.1 Khz. I'd like to see some ABX test results to prove it."
this was the initial idea:
the diferences in quality between 44k and 48k,
--->not to re encode 44k to 48k that is no sense!
and:
"Stereo is _not_ better than joint stereo. Joint stereo will always have more quality, the encoder can be more efficient with more bits shared between....."
another mistake....
stereo and/or dual channell is better but you don't trust me!
i,Boulder,DDogg and the audio faq from Doom9
and the links posted was clear about it!
...and wrote:
"r6d2 is correct."
in what?
cos r6d2 wrote:
" but the idea was to see if the distortion was audible"
the question is that 48 sounds better and not if
with 44 the distortion(if have any) was audible!
or
that i could "cheat"?
this is not the idea!

i confirm that 48k sounds better than 44k
--->if:
use 192 bitrate for musicals (128 for conversations).
use high quality resampler such as SSRC cos
lame is(tunned)better for 44k,not for 48k!
use 2 pass mode (like in headac3he).
use stereo or dual channel!
normalise at 98% cos loud volumes not means quality!
than the quality will be diferent!

--->was used 2lame in your samples,
no matter what you did,you loose quality.
file1 and file2 have diferents volumes,
the left channel have more than 3db than right channel,
and the sound clip(saturation in left channel)!
i can't believe that the source have this diferences
your encodes deserve more "care".

well,i send the mail with my "impressions"
you can't trust in what i post here but
i trust that you did your samples like you wrote
cos we want good quality...
was really good your tests cos just a few people want quality,
see the number of posts in the thread!
no matter what file was re encoded,
using my 3 amplifiers(1 prologic) and 10 speakers
i can feel(clear) that 48 is better than 44k(without re encode to 48k).
this i can confirm!
but re encoded 44-->48 in 128, with 2lame
and with 3db pan diferences,cliping and volume diferences
...i have my doubts!
:confused: :p

please,post the "general" results and who send mails to you!

thanks.

:)

[KpeX]

@jorel

Q15 of the audio FAQ was recently added, not edited.

The point of this test is to see whether rounding errors introduced in recaluclating sample points when resampling are audible, not to see if one can hear 44.1 khz vs. 48 khz. If you can honestly hear the difference between 44.1 and 48 khz original material, your ears are far above the average human's and the results of this test will not have much relevance for you. What you're saying is that you can hear the difference between sound samples every 1/44100 (.000026) seconds and 1/48000 (.000021) seconds. If you really can, I'm very impressed. If you research more into our recent posts here and in audio encoding you'll see that I had an open mind and did some objective research on the joint stereo issue. If you have a problem with this test and my methods, please disregard the results and don't waste our time.

Edit: also please read these links regarding why 44.1 was chosen for CDs and whether it is enough quality:
http://www.cs.columbia.edu/~hgs/audio/44.1.html
http://www.hydrogenaudio.org/index.php?act=ST&f=1&t=8909
http://www.hydrogenaudio.org/index.php?act=ST&f=1&t=9311

I can tell you that you identified the files correctly. As far as the volume issues, I don't know what you're hearing there, the files were encoded identically. Overall test results will be posted once there is a statistically significant number of results (hopefully soon).

[jorel]

KpeX wrote:
"Q15 of the audio FAQ was recently added, not edited."
yes,you're right....i have the "old" without edition saved (hd)
and thanks for this too!

"you'll see that I had an open mind and..."
yes, i(we all)can see it easily!
:)

the test is really important and is hard
to find who have knowledge and interest
in audio quality like you !
we are only changing opinions and informations
...it can cause some censure but not personal.
don't "bore" with my poor words,
i don't have offensive intentions. :o

best regards!
:)

[KpeX]

jorel:

Thanks for your mature and explanatory post.

@all

Not to get too ahead of myself here, but for the next phase of the test that I'm planning, the filesizes will be significantly higher (.wavs instead of .mp2), and I'll need someone to host the files as my free webspace sucks. Requirements would be hosting of a 30-50 mb zip file with decent speed. Any volunteers feel free to PM me.

[KpeX]

Well this thread seems pretty quiet so I guess I'll wrap things up. Out of five testers not including myself only one correctly identified file1 as the resampled file. interesting to note is that several almost leaned towards preferring file1, the resampled version. Although five is far from statistical significance it is something to learn. It doesn't seem like the interest in testing is too high here but a big thank-you to all who participated.

[r6d2]

Originally posted by KpeX
Out of five testers not including myself only one correctly identified file1 as the resampled file. interesting to note is that several almost leaned towards file2, the resampled version.
Well, I don't quite understand which was the resampled one, since you seem to refer to both by the name of resampled. I looked at your original question and then checked the e-mail I sent you, and I'm not sure if I expressed myself in a clear way:
Hard to tell, really. My eras [typo, ears] are not what they were sometime ago. But if a choice has to be made, I’d rather prefer file1.
What I meant was that I'd prefer to use the file1 version for the encode. Would you clarify which one was the 48->44->48?

[KpeX]

Wow, that's a serious typo on my part. Thanks for pointing that one out r6d2. I fixed my post now, file1.mp2 is the one that went through the 44.1>48>44.1 cycle. I understood what you said in your email fine.

[r6d2]

Well, I think the name of the only THX enabled ears owner deserves to be revealed, don't you agree? ;)

[KpeX]

http://forum.doom9.org/member.php?s=&action=getinfo&userid=14118

[r6d2]

Congrats jorel! :cool:

[jorel]

Originally posted by r6d2
Congrats jorel! :cool:

thanks r6d2 and KpeX!

but so sad cos just a few people is searching good audio quality,
like i "told" you KpeX,is hard to find who like good things!

and i ever say(of topic but interesting):
"volume not means quality"
i can't wrote good explanations like in this links:

http://www.mindspring.com/~mrichter/
and
http://www.loudnessrace.net/
thanks to LTR12101B from dvdrbase forum for this links!

i have old cds(aad) with better sound than news cds(ddd),
metalica (load and reload) and megadeth(rude awekening)
sounds horrible...."everything" is 0(zero) db.
when you listen more than 15 minutes,the ears are "tired".

:)

[pacohaas]

also, identifying one audio sample doesn't say much (statistically). To truely "prove the worthiness" of his ears, he'd have to identify a large percentage of many samples(which i don't doubt is possible). But a doubter here could always just say it was a lucky guess. Maybe doom9 can announce the next test on the News page and you could notify HA about the post as well. I think it's also important that people DON'T post their results or thoughts of any kind regarding the test files on the thread, and if you(KpeX) or I see such a post, we should edit it away and PM the user right away so as to not influence the rest of the testers.

oh, and thanks for your work ;)

[KpeX]

@pacohaas

I don't think there's enough interest in this community for audio listening tests. That's one reason I didn't do anything too extensive with this test - I wanted to test the waters, and it seems most people would rather trust their own ears.

@ all

if you'd be interested in more mp2 listening tests (bitrate/encoders/etc.) post here.

[r6d2]

Originally posted by KpeX
it seems most people would rather trust their own ears.
Let me rephrase: most people here are more concerned with video quality or just have standard stereo ears and not THX enabled ones like jorel. :D:D:D