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: Another Joint Stereo Discussion (Read 38497 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Another Joint Stereo Discussion

Reply #100
The L/R <-> M/S transformation may be performed losslessly, with the proper math.  This is why it's used adaptively in every lossless codec I know, as it's an efficiency boost in almost every case.

To put away any doubt about the process itself, and its possible efficiency boost:

Code: [Select]
D:\Music\Z Unsorted>wavpack -f -j0 -m "neon.wav" "neonlr.wv"

WAVPACK  Hybrid Lossless Wavefile Compressor  Win32 Version 4.2  2005-04-02
Copyright (c) 1998 - 2005 Conifer Software.  All Rights Reserved.

original md5 signature: fbae7db045b46a82c242dab87f04210a
created neonlr.wv in 3.27 secs (lossless, 28.20%)

D:\Music\Z Unsorted>wavpack -f -j1 -m "neon.wav" "neonjs.wv"

WAVPACK  Hybrid Lossless Wavefile Compressor  Win32 Version 4.2  2005-04-02
Copyright (c) 1998 - 2005 Conifer Software.  All Rights Reserved.

original md5 signature: fbae7db045b46a82c242dab87f04210a
created neonjs.wv in 3.16 secs (lossless, 29.42%)

D:\Music\Z Unsorted>wvunpack -m -v *.wv

WVUNPACK  Hybrid Lossless Wavefile Decompressor  Win32 Version 4.2  2005-04-02
Copyright (c) 1998 - 2005 Conifer Software.  All Rights Reserved.


neonjs.wv:
original md5:  fbae7db045b46a82c242dab87f04210a
unpacked md5:  fbae7db045b46a82c242dab87f04210a
verified neonjs.wv in 2.28 secs (lossless, 29.42%)

neonlr.wv:
original md5:  fbae7db045b46a82c242dab87f04210a
unpacked md5:  fbae7db045b46a82c242dab87f04210a
verified neonlr.wv in 2.25 secs (lossless, 28.20%)

**** 2 files successfully processed ****

D:\Music\Z Unsorted>dir neon*
Volume in drive D is 114.0
Volume Serial Number is A8F9-887B

Directory of D:\Music\Z Unsorted

2005.06.03  02:09 PM        46,233,308 neon.wav
2005.06.03  02:24 PM        32,629,700 neonjs.wv
2005.06.03  02:24 PM        33,193,642 neonlr.wv
              3 File(s)    112,056,650 bytes
              0 Dir(s)  20,989,685,760 bytes free


Just an encoding of John Mayer's "Neon" from Room for Squares, with JS forced off, then enabled (which is the default, but I forced it on to remove doubt about that fact).  This codec is indeed purely lossless, and obviously in this sample, the use of JS was also lossless, and saved 0.2% on the total bitrate.  This will obviously differ for each file, but: it is indeed purely lossless - both md5s match, and you can feel free to check out the code in "pack.c" and try it yourself.

Edit: I apologize for the long post, but it seems like a pretty easy, convincing case of lossless JS usage, and establishing the process as lossless seems like it would help clear some things up.

Another Joint Stereo Discussion

Reply #101
Quote
@Jebus
"Well guys, i'm a purist, i don't mind size, and i don't want 1 bit error, so i use 320 L/R. This configuration is the most preserving". And i would be in the true saying this.


Argh! No! That's my point! If that was the case, EVERYONE would use L/R stereo! 320kbps, while it might seem like a lot, is NOT. It is NOT enough to encode everything with everything preserved! If you throw away bitrate towards L/R stereo, you LOOSE bitrate that would have been better spent encoding other stuff. So in short, you are REDUCING the quality by forcing L/R.

I have a theory why you newbies always argue L/R stereo: It is about the only bullet-point feature you readily understand. "Real stereo or fake stereo? I want REAL stereo!" No offense, but if you had a better grasp of lossy coding then you would NOT be having this conversation with us. Have you noticed not a single developer has taken your side?

Another Joint Stereo Discussion

Reply #102
Weren't vinyl records recorded in M/S stereo, anyway? Therefore... Joint Stereo sounds "warmer" than normal stereo! 

Another Joint Stereo Discussion

Reply #103
Quote
I can say it because i'm a drummer and i can hear the absence
[a href="index.php?act=findpost&pid=302573"][{POST_SNAPBACK}][/a]

Quote
what's happen if the algorithm says that 2 channels are equal (resulting in full MID and 0 SIDE), while they are not *entirely*?
[a href="index.php?act=findpost&pid=302573"][{POST_SNAPBACK}][/a]

Quote
it's evident in this scenario that J/S is faulty.
[a href="index.php?act=findpost&pid=302573"][{POST_SNAPBACK}][/a]

Quote
in a PRECISE HIGH QUALITY SCENARIO it could be safe to encode all in L/R
[a href="index.php?act=findpost&pid=302573"][{POST_SNAPBACK}][/a]

Quote
- For a super secure paranoid encoding it's better use always L/R stereo
[a href="index.php?act=findpost&pid=302802"][{POST_SNAPBACK}][/a]

Quote
So there's no point in "weighing". Because a perfect stereo preservation can contribute to the "clean sounding" of an mp3 and so to its overall quality.
[a href="index.php?act=findpost&pid=302875"][{POST_SNAPBACK}][/a]

Quote
In my opinion M and S have 50% and 50% importance, to preserve a trasparent sound.
[a href="index.php?act=findpost&pid=303193"][{POST_SNAPBACK}][/a]

Quote
People who think about obtaing the EXACTLY quality at the 1/7 of size are simply out of the world
[a href="index.php?act=findpost&pid=303193"][{POST_SNAPBACK}][/a]

Quote
Your post doesn't make any single sense. Re-read and understand better the J/S working
[a href="index.php?act=findpost&pid=303193"][{POST_SNAPBACK}][/a]


I call troll.

Another Joint Stereo Discussion

Reply #104
yep, no one is this dense. Move along people, nothing to see here.

Another Joint Stereo Discussion

Reply #105
I don't think i'm a troll at all.
When i start to read this thread i want only to test the two samples posted, and say my opinion. But as the discussion went by, i started to read here and there in order to understand better the real working and real problems of either L/R or M/S, noise on separate chans, 1bit lose, s-starving and so on, then i take my conclusion, basing on pros and cons of each method.
And yes i've said most-preserving not best-sounding , 'cause i know very well the fact that one thing is preserving, and other thing is how it sounds in the global at your hear (infact i've precised it was only a kind of joke phrase).
Anyway all is stated in the phrase someone said : "Have you noticed not a single developer has taken your side?". Why a developer would have taken my side? I'm not doing a war, nor criticize one method against one other. I'm only expressing my totally legitimate doubts.
But here, as sometimes happens, seems to be existed "the holy church of Joint Stereo", where all is undubitable dogma, Joint Stereo is the best on all and forever, and no one can speak about it.
And again yes, i started discussing to you because i HAVE a grasp of what a lossy can be (or would be) or can't be. But unfortunately here it seems to be the holy church.
Ah, a little notation for all you M/S fanatics (at this point i think so), just another thought to demonstrate how much wrong is your way to look at things : i'm glad that exists a form of completely lossless M/S notation (thanks Ariakis), 'cause i think M/S is a very interesting way of looking at a stereo signal.
Sadly bye.

Another Joint Stereo Discussion

Reply #106
*slaps forhead*

ALL m/s stereo works like that!

Another Joint Stereo Discussion

Reply #107
More preserving means keeping more of the original data. Since lossless compression is keeping the original data, any extra losses incurred by not encoding in MS make LR stereo less preserving. No one's saying Joint Stereo is the best for everything - that's why LAME switches between them. Forcing LR stereo makes for bigger and/or lower quality files in all but the most extreme circumstances, but forcing MS stereo on most files would usually make slightly bigger and/or low quality files too. That's the reason the presets switch between them, depending on which is more appropriate...

Another Joint Stereo Discussion

Reply #108
unfortunately nobody dealt with my previous post so far, which proved, that inside of joint stereo m/s coding are shades of grey, from black to white. From Lossless joint stereo, the --nssafejoint parameter of lame preset 320/insane, the msfix values, in MPC the different --ms (var) values.......

Another Joint Stereo Discussion

Reply #109
Quote
More preserving means keeping more of the original data. Since lossless compression is keeping the original data, any extra losses incurred by not encoding in MS make LR stereo less preserving. No one's saying Joint Stereo is the best for everything - that's why LAME switches between them. Forcing LR stereo makes for bigger and/or lower quality files in all but the most extreme circumstances, but forcing MS stereo on most files would usually make slightly bigger and/or low quality files too. That's the reason the presets switch between them, depending on which is more appropriate...
[a href="index.php?act=findpost&pid=303385"][{POST_SNAPBACK}][/a]

You're making me kinda smile
Well, why L/R is more preserving (and i repeat not necessary best sounding)? I demostrate you.
Take a wav and use on it the maximum allowed L/R setting that is CBR 320kbits. 320 means that every second is quantized in 320000 bits. CBR means that EVERY seconds is quantized in 320000 bits. So we have in L/R CBR 320, that the wav was trasformed in (320000 * n seconds) bits.
Now try M/S encoding. In a J/S context i don't know if it's possible to quantize M & S at 320kbit, but we assume for our extreme example that is possibile. Let's go.
We have our wav, we have to trasform into a M/S notation before compressing, and we know that M/S loses 1bit for each trasformation. So, assuming 320kb quantitation is possible we would have the wav trasformed in ((320000 - 1) * n seconds) bits.
And this just would be suffice to declare Stereo Mode more bits preserving (and repeat for the n^ time, not necessary best sounding).
But back into a normal context, M/S i think  is normally used in combination with a VBR preset, so in a normal context we have (((56k < x > 320k)-1) * n seconds)bits which is even less than the extreme hypotetical ((320000 - 1) * n seconds) bits.

L/R is the most preserving bits and there can be no question about it .

Another Joint Stereo Discussion

Reply #110
Give it up guys. That's a troll and no question about it.

Another Joint Stereo Discussion

Reply #111
Its good to be happy 3ngel, its not good for you to post this nonsense in Hydrogen Audio. You take a binary digit from here and pop it into an unrelated equation without a care for logical continuity. You are as an astrologer arguing about the spin rate of neutron stars. You have intellegence but are applying it to a fantasy, that you can correct workings of intricate mechanisms which you wont accept advice on how to safely lift the lid off nevermind re-engineer.
This is not a fantasy roll playing board, despite the appearance of the trollslayers.
Lurk for a while and listen, it shouldnt take a guy of your insight and mental agility too long to learn something if you are truely curious about the subjects.

Good luck and all the best, keep spamming this thread with confident ignorance and you may leave it in the traditional untellable newbie's firedance
no conscience > no custom

Another Joint Stereo Discussion

Reply #112
@ChiGung
Sure, pheraps i'm saying nothing but bul***it, but obviously my will is to learn and discuss possibly without saying false things. If you all are right, and i'm saying nothing but nonsense, well this means i've not understood not a single bit of how M/S works (but i don't think so). In any case, i will not post anymore and i'll lurk and listen in order, as you say, to put in doubts again all my "convinction".
But let me say one thing : i don't know if you are a dev or what, but until now no one has done a clear demostration i'm so wrong. All the things i read are assertions based on "wtf is so obvious! everyone say that", but the fact that everyone say that, doesn't mean it's true.
In the end, my intention was not to create useless discussions, but all the opposite, and i thank you and the others for the pieces of knowledge i've learnt from you.
Time for me will say if i'm wrong or right.

Another Joint Stereo Discussion

Reply #113
Quote
........
In the end, my intention was not to create useless discussions, but all the opposite, and i thank you and the others for the pieces of knowledge i've learnt from you.
Time for me will say if i'm wrong or right.

I appreciate your mild response 3ngel, Im tempted but wont try again to convince you of the veracity of previous reasoning that M/S is not fundamentaly inferior. Briefly, the perfect encoder will benefit from being able to choose between any available notations which the numbers at that instant happen to favour... and leave that rounding bit out of your reasoning because you repeatedly misplaced it.

Ive made some contentious suggestions myself in previous posts so will be lurking too, but I know not to hold my breath here - beggars for attention cant be choosers
The closest to the rockface, dont speak much at all.
 
Im not a dev here, but am deving in private, cowardly - I cant talk about my most promising project until ive enough of a headstart on it to share any findings.

cya l8r then, atb

[span style='font-size:7pt;line-height:100%']edit: come to think of it, i could just be a cranky old fart[/span]
no conscience > no custom

Another Joint Stereo Discussion

Reply #114
Quote
Now try M/S encoding. In a J/S context i don't know if it's possible to quantize M & S at 320kbit, but we assume for our extreme example that is possibile. Let's go.
We have our wav, we have to trasform into a M/S notation before compressing, and we know that M/S loses 1bit for each trasformation. So, assuming 320kb quantitation is possible we would have the wav trasformed in ((320000 - 1) * n seconds) bits.
...
L/R is the most preserving bits and there can be no question about it .
[a href="index.php?act=findpost&pid=303408"][{POST_SNAPBACK}][/a]

Read Ariakis' post - M/S does not lose 1 bit per transformation. If they're both going to 320kbps then, purely on a bit level, each is preserving just 320kbps of the original 1411 - no more, no less. However, in many cases the 320kbps that M/S preserves actually represents more data and sound than the 320kbps that the L/R preserves, because it's undergone a further lossless compression beforehand. That's not what I think, that is in basic terms how joint stereo is implemented in LAME... beyond stating that over and over again there's not much useful I can bring to the topic, it seems...

Another Joint Stereo Discussion

Reply #115
Quote
We have our wav, we have to trasform into a M/S notation before compressing, and we know that M/S loses 1bit for each trasformation.


Not necessarily. If the M/S values are stored as long integers (32 bits), there is no loss.

Quote
So, assuming 320kb quantitation is possible we would have the wav trasformed in ((320000 - 1) * n seconds) bits.


1 bit would be lost for each conversion, and there are 88200 conversions per second (encoding plus decoding). But anyway your demonstration is wrong. You do not convert 1411 kb into 320 kb in L/R or 320 k - 1 b in M/S, you would actually convert 1411 kb into 320 kb in L/R, and 1411 k -1 b into 320 kb in M/S

Otherwise, since there are 88200 bits lost per second, it would not be possible to work below 88 kbps, because we would have a negative number of bits left

Another Joint Stereo Discussion

Reply #116
@tgoose
I think that Ariakis post was referring to "WAVPACK  Hybrid Lossless" which demonstrate that exists a M/S notation conversion lossless, but not that Lame do it in a lossless way. Moreover I remember was Chigung who confirmed that there was a 1 bit lossy (even if he says it's a last importance bit, but even if last importace i say it's present ) in a Lame context.

@Pio2001
Quote
Not necessarily. If the M/S values are stored as long integers (32 bits), there is no loss.

Yes, this can be right (i think), but that's not the case of lame (as i've read until now)

Quote
But anyway your demonstration is wrong. You do not convert 1411 kb into 320 kb in L/R or 320 k - 1 b in M/S, you would actually convert 1411 kb into 320 kb in L/R, and 1411 k -1 b into 320 kb in M/S

Yes, you're right, in the rush i did a little error

In the end i think that if it was not for the 1 bit lose, the two methods (L/R & M/S) could be considered from a bit preserving point of view equal (and theorically before compression they are 'cause they are only notation change), while from a psyco-acu point of view i think that in a certain music context (aka genres), at a given bitrate pheraps one can be more "sound-pleasing" than the other and viceversa, because of the pros and cons of each method.

Another Joint Stereo Discussion

Reply #117
Quote
@tgoose
I think that Ariakis post was referring to "WAVPACK  Hybrid Lossless" which demonstrate that exists a M/S notation conversion lossless, but not that Lame do it in a lossless way. Moreover I remember was Chigung who confirmed that there was a 1 bit lossy (even if he says it's a last importance bit, but even if last importace i say it's present ) in a Lame context.

@Pio2001
Quote
Not necessarily. If the M/S values are stored as long integers (32 bits), there is no loss.

Yes, this can be right (i think), but that's not the case of lame (as i've read until now)
[a href="index.php?act=findpost&pid=304237"][{POST_SNAPBACK}][/a]


Quote
*slaps forehead*

ALL m/s stereo works like that!
[a href="index.php?act=findpost&pid=303342"][{POST_SNAPBACK}][/a]

Satisfied? 

Another Joint Stereo Discussion

Reply #118
Quote
@Pio2001
Quote
Not necessarily. If the M/S values are stored as long integers (32 bits), there is no loss.

Yes, this can be right (i think), but that's not the case of lame (as i've read until now)

lame uses a floating point representation, not integer.

frankly, if you're worried about such a simple transformation, you should avoid using any lossy compression scheme.

Another Joint Stereo Discussion

Reply #119
It isn't worth to discuss with this 3angel.
Maybe somebody tells him, that eg. the hq HDCD format throws 1 bit of 16 away, ie. the 16th bit produces only noise, if hdcd played via normal non-hdcd-capable machines.
But nobody of the "audiophiles" have complained so far...
and here he is discussing a rounding error...

well, regarding joint stereo / m/s encoding,
I would be interested in an explanation about the way, what the msfix switches do in particular, nssafejoint, or the various --ms xy switches in MPC.
All belong to m/s encoding, but the results are different "qualities" regarding m/s or "stereo-separation" ?

Another Joint Stereo Discussion

Reply #120
Quote
Sure, pheraps i'm saying nothing but bul***it,


You are.

Quote
If you all are right, and i'm saying nothing but nonsense, well this means i've not understood not a single bit of how M/S works (but i don't think so).


You haven't.

Quote
But let me say one thing : i don't know if you are a dev or what, but until now no one has done a clear demostration i'm so wrong.


You haven't shown that you're right either.  And given that you don't understand what you're talking about, I don't see why anyone is obligated to prove you wrong.

Another Joint Stereo Discussion

Reply #121
Wow! This was a great discussion. Thanks!

Well, except for some of the more inane rambling that got mixed in around the middle/end. But I digress.  :-)

I have LONG held that there is a difference between "stereo" and "joint stereo" and that I can hear that difference. However, when I first posted to that effect, I was flamed up one side and down the other. I still bear the scars of that encounter.

*ouch*

I'm a professional musician and an audio engineer...having worked in semi-pro and pro recording studios for years. I work in a business where tonal subtleties are stock-in-trade. Boutique tube amps are sought for their "tone". Solid state amps try to "model" that same elusive "tube sound". It's the nature of my work to listen critically. And I am very sensitive to phase and other issues which effect stereo separation. My EARS have always been able to hear a difference between a file encoded in stereo versus a file encoded in joint stereo. I realize this kind of assertion drives the objectivists aound these parts MAD. But I cannot prove mathematically what can only be HEARD. Perceptual differences are just that...perceptual. I have encoded the same song countless times as follows.

--alt-preset standard

--alt-preset standard -ms

Then I have ABX'ed between the two files in Foobar. And I hear a difference. Is one "better" sounding than the other? That's an entirely different question and one which I am not sure I could answer to anyone's satisfaction other than my own. But I do hear a difference.

Can I show you that difference on a chart or graph? Or can I convince you to hear the same difference that I hear? Even if I post a hundred samples? Probably not. A friend of mine is a professional audiologist. And he tells me that the differences between any two individuals hearing is FAR greater than the subtle differences between two different pieces of electronic equipment. And I believe him. No two people hear exactly alike. And no two people perceive sound exactly alike.

But, that does NOT invalidate subjective listening. Just because an instrument cannot (currently) measure a difference, does that mean the perceived difference does not exist? That would be to presume that the state of our current instrumentation is perfect and cannot be improved upon. That is the height of professional arrogance. And we know from scientific experience that it is not true. In ten years, we will be able to measure and quantify things we didn't even know existed in 2005. Science WILL catch up with art one day. *joke* Peace to all objectivists!

I know we cannot settle the objectivist/subjectivist debate that rages in the audio community here on our little forum. But I would like to see both camps start to listen to one another a little more. It makes for much more interesting discussion when we all lay aside our biases. For example, I am NOT saying that the much beloved "joint stereo" setting is inferior. I merely contend that it is different from -ms (and other various "stereo" settings). And I would really like to see more work done toward quantifying that difference.

Wouldn't it be great if we could set up a mass ABX test hands on? A bunch of us in the same room listening to different material and discussing the perceived differences? Man...I can dream can't I? :-)

Another Joint Stereo Discussion

Reply #122
Sorry, quoting doesn't work.
Quotes are in italic.

My EARS have always been able to hear a difference between a file encoded in stereo versus a file encoded in joint stereo. I realize this kind of assertion drives the objectivists aound these parts MAD.

Not necessarily. The point is that it violates one of the Terms of Service of this board. Number 8, to be precise.

But I cannot prove mathematically what can only be HEARD.

Yes, you can. An ABX test performed according to this rules ( http://www.hydrogenaudio.org/forums/index....howtopic=16295& ) can be considered as a mathematical proof.

Then I have ABX'ed between the two files in Foobar. And I hear a difference.

Well, in this case, can you tell how many attempts you did make before getting a good score, how much times did you recognize the good file, how much trials you performed, if you were looking at your score during the test, and if you had decided the number of trials before starting ?

Is one "better" sounding than the other? That's an entirely different question and one which I am not sure I could answer to anyone's satisfaction other than my own.

That's perfectly right. Blind tests can show that someone hears a difference, they can (in ABC/HR) show which sound the listeners prefers under a blind comparison, but can't tell which one is better.

Can I show you that difference on a chart or graph?

Yes, the number of ABX successes over the total number of trials allows to calculate p, the probability that your success doesn't come from what you hear. The more trials you decide to perform prior to the test, the smaller is p, granted that you can hear a difference of course.

Or can I convince you to hear the same difference that I hear?

No. In this forum, we restrict discussions to the scientific approach only. You can just show us the right direction by describing what allowed you to succeed the ABX test (after having explained your ABX methodology), so that others can try the same sample.

Just because an instrument cannot (currently) measure a difference, does that mean the perceived difference does not exist?

Good question. Finding an audible difference (under blind listening conditions, of course) that can not be measured would give an answer, but to my knowledge, this has never been done.

That would be to presume that the state of our current instrumentation is perfect and cannot be improved upon

You're mixing up audible differences with existing differences.
Think about temperature. All perceiveable differences are measurable with a thermometer. Thus if a difference is not measurable, it means that it cannot be perceived. However it doesn't mean that thermometers are perfect. It just means that they are more sensitive than the human body.

I know we cannot settle the objectivist/subjectivist debate that rages in the audio community here on our little forum.

If you use the words "objectivist" and "subjectivist" in their usual meaning in audio (controlled blind tests versus no controlled blind tests), then it has already been settled in this forum. Only objectivist discussions are allowed. Subjectivism is considered as off-topic and removed from the discussions. That's just the way the forum have been created. We don't pretend objectivism to be superior. That's just the same thing as changing rooms. There are changing rooms for women and changing rooms for men. The fact that men are not allowed in women chaning rooms doesn't mean that women think that men are inferior. It means that this is a changing room for women.
Here, we are in a forum for objectivists.

But I would like to see both camps start to listen to one another a little more. It makes for much more interesting discussion when we all lay aside our biases.

In this perspective, I think that criticism about blind tests can be constructive.

Wouldn't it be great if we could set up a mass ABX test hands on? A bunch of us in the same room listening to different material and discussing the perceived differences? Man...I can dream can't I? :-)

Roberto Amorin did setup mass ABX tests http://www.rjamorim.com/test/
Different codecs were tested, and the results were always discussed here.