Hydrogenaudio Forums

Lossy Audio Compression => MP3 => MP3 - Tech => Topic started by: crazyman on 14 January, 2005, 08:46:16 AM

Title: Any listening test Stereo vs Joint stereo in mp3?
Post by: crazyman on 14 January, 2005, 08:46:16 AM
Hi all,

in many discussions I read about mp3 via Lame encoder the result is that the best solution is joint stereo. I believe in that as the developers who I highly esteem say so. But is this statement supported also by more independent listening tests? Especially I would be interested in comparison between --preset standard (joint stereo) and dtto with stereo mode.

Can the result in general depend also on the Lame version? From the tests I could read usually still the old good Lame 3.90.3 comes out as the winner.

Cheers,

Crazyman
Title: Any listening test Stereo vs Joint stereo in mp3?
Post by: beto on 14 January, 2005, 08:51:01 AM
This makes no sense, since joint stereo is just a way to encode stereo losslessly.

There should be no audible difference between them AFAIK....
Title: Any listening test Stereo vs Joint stereo in mp3?
Post by: mithrandir on 14 January, 2005, 10:38:51 AM
Quote
This makes no sense, since joint stereo is just a way to encode stereo losslessly.

There should be no audible difference between them AFAIK....
[a href="index.php?act=findpost&pid=265290"][{POST_SNAPBACK}][/a]

No, joint stereo is lossy. That's what directed the Scene to use full stereo mode (not that they really know what they are doing). But joint stereo can be objectively transparent if implemented properly...and it can save you a lot of bits. There is generally no need to use full stereo in LAME.
Title: Any listening test Stereo vs Joint stereo in mp3?
Post by: SirGrey on 14 January, 2005, 10:48:36 AM
>>No, joint stereo is lossy.
Wrong.
joint-stereo is lossless, intensity stereo is lossy.

The only thing about it, that is testable is switching algorhytm - i.e. what representation saving more bits.
EDIT:
Quote
beto: This makes no sense,

Agree 
Title: Any listening test Stereo vs Joint stereo in mp3?
Post by: Otto42 on 14 January, 2005, 10:50:59 AM
Quote
No, joint stereo is lossy. [a href="index.php?act=findpost&pid=265313"][{POST_SNAPBACK}][/a]

Depends on what you're talking about. As I understood it, there's two ways to do Joint Stereo. Mid/Side Stereo is lossless. Intensity Stereo is lossy. But intensity stereo is only used at low bitrates (96k and less) and so Joint Stereo is lossless at normal bitrates used for audio anyway.

So in other words, Joint Stereo can only improve quality by offering another possible data set to compress and giving a possible higher quality. Using Stereo Only makes no sense because that's just eliminating the possibility of using M/S encoding.
Title: Any listening test Stereo vs Joint stereo in mp3?
Post by: SirGrey on 14 January, 2005, 10:53:31 AM
Addenium:
Lots of the confusion goes from Fhg.
They usually call ms stereo as joint stereo and intensity stereo as ms-is stereo. 
Title: Any listening test Stereo vs Joint stereo in mp3?
Post by: Gray_Wolf on 14 January, 2005, 11:10:22 AM
This is incredible 

Many people in present moment believe that Lame CBR 192kbps - Stereo mode is the best solution   
Title: Any listening test Stereo vs Joint stereo in mp3?
Post by: Jojo on 14 January, 2005, 11:53:30 AM
The bottom line is that Joint-Stereo used in LAME is lossless. Therefore, it will either improve quality or reduce filesize. So, Joint-Stereo + LAME is a must! The Stereo mode in LAME should be removed since there is no use for this anymore...
Title: Any listening test Stereo vs Joint stereo in mp3?
Post by: dev0 on 14 January, 2005, 12:42:50 PM
Quote
No, joint stereo is lossy. That's what directed the Scene to use full stereo mode (not that they really know what they are doing). But joint stereo can be objectively transparent if implemented properly...and it can save you a lot of bits. There is generally no need to use full stereo in LAME.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=265313")


No, Joint Stereo is lossless concerning stereo information. Some scene groups are using 'full stereo' VBR, because using a Dolby Pro Logic or similliar surround DSP will unmask JS artifacts, which are masked under normal (=stereo) playback conditions.
For more information see this thread:
[a href="http://www.hydrogenaudio.org/forums/index.php?showtopic=12004]http://www.hydrogenaudio.org/forums/index....showtopic=12004[/url]
Title: Any listening test Stereo vs Joint stereo in mp3?
Post by: Mark7 on 14 January, 2005, 01:04:18 PM
Quote
Quote
No, joint stereo is lossy. That's what directed the Scene to use full stereo mode (not that they really know what they are doing). But joint stereo can be objectively transparent if implemented properly...and it can save you a lot of bits. There is generally no need to use full stereo in LAME.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=265313")


No, Joint Stereo is lossless concerning stereo information. Some scene groups are using 'full stereo' VBR, because using a Dolby Pro Logic or similliar surround DSP will unmask JS artifacts, which are masked under normal (=stereo) playback conditions.
For more information see this thread:
[a href="http://www.hydrogenaudio.org/forums/index.php?showtopic=12004]http://www.hydrogenaudio.org/forums/index....showtopic=12004[/url]
[a href="index.php?act=findpost&pid=265347"][{POST_SNAPBACK}][/a]


Won't stereo mode destroy the centre channel of DPL suround? Both channels are encoded seperately so there will be a bigger chance that sounds that should the same on each channel are actually be slightly different on each channel and won't go to the centre channel where they belong. (just a theory)

And doesn't DPL more or less play the side channel through the surround speakers? How come there will be artifacts?
Title: Any listening test Stereo vs Joint stereo in mp3?
Post by: mithrandir on 14 January, 2005, 04:44:59 PM
Quote
Quote
No, joint stereo is lossy. That's what directed the Scene to use full stereo mode (not that they really know what they are doing). But joint stereo can be objectively transparent if implemented properly...and it can save you a lot of bits. There is generally no need to use full stereo in LAME.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=265313")


No, Joint Stereo is lossless concerning stereo information. Some scene groups are using 'full stereo' VBR, because using a Dolby Pro Logic or similliar surround DSP will unmask JS artifacts, which are masked under normal (=stereo) playback conditions.
For more information see this thread:
[a href="http://www.hydrogenaudio.org/forums/index.php?showtopic=12004]http://www.hydrogenaudio.org/forums/index....showtopic=12004[/url]
[a href="index.php?act=findpost&pid=265347"][{POST_SNAPBACK}][/a]

I think there is some confusion here...and perhaps it is me. This is what I mean:

If you encode with VBR -V 5, which uses joint stereo not full, it is possible to positively ABX between WAV and MP3 from a stereo perspective (i.e. -V 5 has a smaller soundstage than original thanks to more aggressive use of mid-side frames instead of LR). The variable here is --nsmsfix. Set it low enough (closer to 0) and it would probably be impossible to ABX between MP3 and WAV in terms of stereo performance.

If joint stereo were in fact lossless then why does LAME have to decide between LR and MS frame allocation? It's going lossy in many cases to save bitrate.
Title: Any listening test Stereo vs Joint stereo in mp3?
Post by: Jojo on 14 January, 2005, 07:49:21 PM
Quote
If joint stereo were in fact lossless then why does LAME have to decide between LR and MS frame allocation? It's going lossy in many cases to save bitrate.
[a href="index.php?act=findpost&pid=265405"][{POST_SNAPBACK}][/a]

even if that was the case...Joint-Stereo would be still better than 'Stereo'...I mean, mp3 is a lossy format afterall...however, I don't think that Joint-Stereo used in --preset standard is lossy...
Title: Any listening test Stereo vs Joint stereo in mp3?
Post by: Busemann on 14 January, 2005, 07:55:00 PM
Quote
Quote
If joint stereo were in fact lossless then why does LAME have to decide between LR and MS frame allocation? It's going lossy in many cases to save bitrate.
[a href="index.php?act=findpost&pid=265405"][{POST_SNAPBACK}][/a]

even if that was the case...Joint-Stereo would be still better than 'Stereo'...I mean, mp3 is a lossy format afterall...
[a href="index.php?act=findpost&pid=265437"][{POST_SNAPBACK}][/a]


Huh!?

Doing things that could be messing up the stereo separation is the last thing I'd like to see at high bit-rates
Title: Any listening test Stereo vs Joint stereo in mp3?
Post by: Jojo on 14 January, 2005, 08:03:54 PM
Quote
Quote
Quote
If joint stereo were in fact lossless then why does LAME have to decide between LR and MS frame allocation? It's going lossy in many cases to save bitrate.
[a href="index.php?act=findpost&pid=265405"][{POST_SNAPBACK}][/a]

even if that was the case...Joint-Stereo would be still better than 'Stereo'...I mean, mp3 is a lossy format afterall...
[a href="index.php?act=findpost&pid=265437"][{POST_SNAPBACK}][/a]


Huh!?

Doing things that could be messing up the stereo separation is the last thing I'd like to see at high bit-rates
[a href="index.php?act=findpost&pid=265438"][{POST_SNAPBACK}][/a]

he was talking about -V5, which targets for 130kbps
Title: Any listening test Stereo vs Joint stereo in mp3?
Post by: 2Bdecided on 17 January, 2005, 10:34:28 AM
The stereo to joint stereo (i.e. L+R>M+S) transform is lossless and perfectly reversible. However, the channels are then lossily coded.

If you force an encoder to (a) stay in L+R mode (i.e. "discrete" or "true" stereo), and then (b) force it into a bitrate area where it will add audible distortion, the stereo image will be preserved but you'll have audible quantisation noise. If anything, the addition of poorly-correlated noise to both channels may make the stereo sound stage sound wider - artificially and quite nastily of course. The centre image could be damaged.

If you force an encoder to (a) stay in M+S mode (i.e. one of the options available in "joint" stereo), and then (b) force it into a bitrate area where it will add audible distortion, the stereo image will start to collapse and you may have audible quantisation noise, but it will be less obvious.

The point of VBR is to use as many bits are as necessary for a transparent result, and the point of joint stereo is to give the option of encoding M+S (or L+R!). You're only going to hear a difference between discrete and joint stereo if something goes wrong. Which it rarely does with lame 3.90.3 aps. What's more, of all the thousands of tracks that must have been encoded with aps, and the handful of tracks which are not transparent, no one has suggested that any of these are improved by forcing discrete stereo.

I'd encourage people to look at that thread...
http://www.hydrogenaudio.org/forums/index....showtopic=12004 (http://www.hydrogenaudio.org/forums/index.php?showtopic=12004)
...to see that forcing stereo may not be the answer - even when using a DPL or DPLII decoder.

Was this issue ever resolved?

Cheers,
David.
Title: Any listening test Stereo vs Joint stereo in mp3?
Post by: crazyman on 19 January, 2005, 03:25:46 AM
Quote
The stereo to joint stereo (i.e. L+R>M+S) transform is lossless and perfectly reversible. However, the channels are then lossily coded.

If you force an encoder to (a) stay in L+R mode (i.e. "discrete" or "true" stereo), and then (b) force it into a bitrate area where it will add audible distortion, the stereo image will be preserved but you'll have audible quantisation noise. If anything, the addition of poorly-correlated noise to both channels may make the stereo sound stage sound wider - artificially and quite nastily of course. The centre image could be damaged.

If you force an encoder to (a) stay in M+S mode (i.e. one of the options available in "joint" stereo), and then (b) force it into a bitrate area where it will add audible distortion, the stereo image will start to collapse and you may have audible quantisation noise, but it will be less obvious.

The point of VBR is to use as many bits are as necessary for a transparent result, and the point of joint stereo is to give the option of encoding M+S (or L+R!). You're only going to hear a difference between discrete and joint stereo if something goes wrong. Which it rarely does with lame 3.90.3 aps. What's more, of all the thousands of tracks that must have been encoded with aps, and the handful of tracks which are not transparent, no one has suggested that any of these are improved by forcing discrete stereo.

I'd encourage people to look at that thread...
http://www.hydrogenaudio.org/forums/index....showtopic=12004 (http://www.hydrogenaudio.org/forums/index.php?showtopic=12004)
...to see that forcing stereo may not be the answer - even when using a DPL or DPLII decoder.

Was this issue ever resolved?

Cheers,
David.
[a href="index.php?act=findpost&pid=266029"][{POST_SNAPBACK}][/a]


David,

many thanks. BTW, mp3 gain is also one of necessary tools for me.
Rgds,

Crzmn
Title: Any listening test Stereo vs Joint stereo in mp3?
Post by: crazyman on 19 January, 2005, 03:30:47 AM
Quote
This is incredible  

Many people in present moment believe that Lame CBR 192kbps - Stereo mode is the best solution   
[a href="index.php?act=findpost&pid=265325"][{POST_SNAPBACK}][/a]


Quite agree. Even some good rippers (f.i. CDEx) in help file say that for internal encoder, stereo is the recommended option ...

Rgds,

Crzmn
Title: Any listening test Stereo vs Joint stereo in mp3?
Post by: Busemann on 19 January, 2005, 05:44:22 AM
Quote
Quite agree. Even some good rippers (f.i. CDEx) in help file say that for internal encoder, stereo is the recommended option ...

Rgds,

Crzmn
[a href="index.php?act=findpost&pid=266494"][{POST_SNAPBACK}][/a]


If it isn't referring to LAME, that isn't necessarily wrong though
Title: Any listening test Stereo vs Joint stereo in mp3?
Post by: mithrandir on 19 January, 2005, 11:52:02 PM
Quote
The stereo to joint stereo (i.e. L+R>M+S) transform is lossless and perfectly reversible. However, the channels are then lossily coded.
[a href="index.php?act=findpost&pid=266029"][{POST_SNAPBACK}][/a]

Thanks for the clarification...this was the source of my confusion.
Title: Any listening test Stereo vs Joint stereo in mp3?
Post by: MugFunky on 20 January, 2005, 12:29:48 AM
guh, why has this myth never been dispelled?

it should be mentioned in the TOS or something.

[edit]  something like "MS joint-stereo is not lossy!  it's simply a way to decorellate data so it can be stored more efficiently".  also mention that the lossy IS stereo mode is not implemented in LAME, and probably never will be.

for a start, just because DPL might bork your stereo image is no reason to pre-bork your mp3s.  dolby pro-logic is not relevant for music that is mixed and mastered as stereo.  and if it were mixed and mastered for DPL, i highly doubt (but cannot prove without an actual DPL decoder) that a preset-standard mp3's transparency will break on it.  so pro-logic is no reason to encode in LR stereo.  DPLII is much more sophisticated than DPL, but again, you'd be stupid to use it on stuff that wasn't meant for it (though apparently it's brilliant on SQ quad records though, having been heavily based on a quad decoder itself).

besides, MS can't be lossy if you're able to put an equals sign between it and LR - maths doesn't work that way.  it's best to leave the decision of where to go LR and where to go MS to lame itself, rather than forcing it.  after all, by forcing LR, you are disabling a feature of the encoder, rather than improving it.
Title: Any listening test Stereo vs Joint stereo in mp3?
Post by: DreamTactix291 on 20 January, 2005, 12:56:15 AM
Quote
guh, why has this myth never been dispelled?

No idea.  Even when I was a newbie after I'd read about the L/R to M/S process I understood it once.  Just two ways to express the same data and letting the encoder pick which one is more efficient is what LAME does in its joint stereo mode.

I agree that somewhere this should be posted where it is obvious that using M/S stereo is not lossy.  If it was certain lossless encoders couldn't use it (I know one does but I can't remember which one)
Title: Any listening test Stereo vs Joint stereo in mp3?
Post by: Gabriel on 20 January, 2005, 03:13:47 AM
The M/S stereo transform itself is lossless.
However, in the context of a lossy encoder, anything leading to some variations in the bits might affect the final loss.
It means that in the context of mp3 encoding, M/S stereo is as lossy as the Huffman coding.
Title: Any listening test Stereo vs Joint stereo in mp3?
Post by: crazyman on 20 January, 2005, 05:29:10 AM
Quote
Quote
Quite agree. Even some good rippers (f.i. CDEx) in help file say that for internal encoder, stereo is the recommended option ...

Rgds,

Crzmn
[a href="index.php?act=findpost&pid=266494"][{POST_SNAPBACK}][/a]


If it isn't referring to LAME, that isn't necessarily wrong though
[a href="index.php?act=findpost&pid=266510"][{POST_SNAPBACK}][/a]


The problem is that it is referring to lame. Also, I used to be a bit confused that is seetings of encoder there were logical lines: Mono, Forced stereo, Joint stereo, Stereo.
Title: Any listening test Stereo vs Joint stereo in mp3?
Post by: 2Bdecided on 21 January, 2005, 06:10:39 AM
Talk about things coming back to bite you...

I'm on a UK media forum, talking about how joint stereo on DAB digital radio is bad (because it's MPEG-1 layer II, which only implements intensity stereo, which is 7-point panned mono above 3, 6, 9, or 12kHz) and someone points me to Hydrogen Audio to learn about why joint stereo isn't really bad at all!

Cheers,
David.
Title: Any listening test Stereo vs Joint stereo in mp3?
Post by: Gabriel on 21 January, 2005, 07:55:55 AM
Well, in your case intensity stereo is a good thing compared to l/r stereo, considering that UK DAB is something like 128kbps Layer II.
Title: Any listening test Stereo vs Joint stereo in mp3?
Post by: Bogalvator on 21 January, 2005, 10:49:37 PM
Quote
Well, in your case intensity stereo is a good thing compared to l/r stereo, considering that UK DAB is something like 128kbps Layer II.
[a href="index.php?act=findpost&pid=267031"][{POST_SNAPBACK}][/a]


Yuck!  I thought it sounded a bit fishy to be honest when I was listening in my mate's car the other day.  Seems I'll save my money rather than investing in DAB car audio (which also doesn't work under bridges etc, just like AM  )

Hey David where is this forum you were on?

I can't understand all the fretting about J/S and it's possible effect on stereo image etc - so what if it was lossy?  The point is does it improve quality, and at higher bitrates, is it ABX-able etc?  Remember that Lame cuts off frequencies above 18000 Hz at -V 2, what's the difference between that and a hypothetical undetectable loss of stereo image?

Doesn't Vorbis implement lossy stereo on q values less than 6?  How well did it do in the last 128kbps test again (-q 4.35 used)?

Perhaps people aren't as sensitive to "stereo image" as they'd like to think!
Title: Any listening test Stereo vs Joint stereo in mp3?
Post by: 2Bdecided on 24 January, 2005, 11:00:33 AM
Quote
Well, in your case intensity stereo is a good thing compared to l/r stereo, considering that UK DAB is something like 128kbps Layer II.
[a href="index.php?act=findpost&pid=267031"][{POST_SNAPBACK}][/a]


Well, yes, but the argument was that joint stereo was OK because (pointers to articles about mp3 joint stereo) - I was trying to argue that, if you _have_ to use joint stereo with mp2, then it can't be transparent - basically because the bitrate is too low.


btw (you people in other countries can have a good laugh about this if you want) - UK DAB digital radio is set to change - if the regulator gets its way, the 128kbps limit will be removed, and replaced by a -2.0 diff grade limit. No idea how this will be measured or enforced, but the broadcasters have reportedly asked for this change so they can move stations down to 112kbps mp2 joint stereo.

Won't that sound nice?   

Some people think that DAB stands for Dead And Buried.

Cheers,
David.
Title: Any listening test Stereo vs Joint stereo in mp3?
Post by: Gabriel on 24 January, 2005, 02:41:44 PM
Quote
but the broadcasters have reportedly asked for this change so they can move stations down to 112kbps mp2 joint stereo.

This will be a good example related to "things to not do"
Title: Any listening test Stereo vs Joint stereo in mp3?
Post by: Ossie on 21 May, 2005, 05:57:57 AM
Hi, I just want to check if the following staement has any validity for music encoded using LAME in Joint Stereo....

Some music contains sounds that are deliberately delayed or phase shifted. Such effect boxes are called "flanger", "phaser" and the like. If you encode such music in 'joint stereo', you will have bad cancelling effects where the high tones appear and disappear all the time, destroying the good original sound. One old example is the accompanying guitar in Paul Simon's "Mrs. Robinson".

Thanks
Title: Any listening test Stereo vs Joint stereo in mp3?
Post by: Pio2001 on 21 May, 2005, 08:01:04 AM
It all depends if the psychoacoustic model can handle them. Some tests need to be done.
You'll have to start from low bitrates in order to try to catch the issue.
If the issue exists, you then need to see if it is still ABXable at high bitrate.
Title: Any listening test Stereo vs Joint stereo in mp3?
Post by: Pio2001 on 21 May, 2005, 08:07:58 AM
Oh, that's not all, you'll have to force mid/side stereo vs left/right stereo for the comparison, then see what stereo mode joint stereo chooses.

I think that a square wave with some pulses would be a good test signal.
Title: Any listening test Stereo vs Joint stereo in mp3?
Post by: mjkng2 on 04 August, 2006, 02:20:03 PM
hey
this is an old article against Joint Stereo: http://harmsy.freeuk.com/mostync/ (http://harmsy.freeuk.com/mostync/)

I don't know, where's the truth. ?
Title: Any listening test Stereo vs Joint stereo in mp3?
Post by: stephanV on 04 August, 2006, 03:16:17 PM
It's not against JS... it just says that an algorithm that works good in theory can be implemented in a a bad way in. That's not something new or spectacular.
Title: Any listening test Stereo vs Joint stereo in mp3?
Post by: Pio2001 on 04 August, 2006, 05:52:23 PM
I've performed two joint stereo listening tests since then.
One with a very special sample : http://www.hydrogenaudio.org/forums/index....st&p=349785 (http://www.hydrogenaudio.org/forums/index.php?s=&showtopic=39514&view=findpost&p=349785)
One with a normal killer sample : http://www.hydrogenaudio.org/forums/index....st&p=404949 (http://www.hydrogenaudio.org/forums/index.php?s=&showtopic=45644&view=findpost&p=404949)
Title: Any listening test Stereo vs Joint stereo in mp3?
Post by: haregoo on 04 August, 2006, 10:57:22 PM
This sample (http://www.fileden.com/files/2006/8/5/156176/uh.flac) is ABXable(16/16) even with -V 0 because of joint stereo.(I confirmed with 3.97b2 and 3.98a6)
JS causes a bit stereo image corruption.

Maybe I made some mistake in a bunch of procedure(encoding, decoding, ABX...etc), so correct me if I'm wrong.
Title: Any listening test Stereo vs Joint stereo in mp3?
Post by: Gabriel on 05 August, 2006, 03:48:37 AM
This sample (http://dl030.filefactory.com/dl/f/bcad21/b/3/h/7088fe1488ed72cc/) is ABXable(16/16) even with -V 0 because of joint stereo.(I confirmed with 3.97b2 and 3.98a6)
JS causes a bit stereo image corruption.

Then please report it in this thread:
http://www.hydrogenaudio.org/forums/index....showtopic=39314 (http://www.hydrogenaudio.org/forums/index.php?showtopic=39314)

Regarding joint stereo or not, have you tryed to ABX with and without joint stereo?
Title: Any listening test Stereo vs Joint stereo in mp3?
Post by: Garf on 05 August, 2006, 04:44:21 AM
hey
this is an old article against Joint Stereo: http://harmsy.freeuk.com/mostync/ (http://harmsy.freeuk.com/mostync/)

I don't know, where's the truth. ?


I think you failed to understand something, because the article is in favor.
Title: Any listening test Stereo vs Joint stereo in mp3?
Post by: haregoo on 05 August, 2006, 09:14:25 AM
Regarding joint stereo or not, have you tryed to ABX with and without joint stereo?

I've done multiple trials with and without -ms.
e.g. 3.97b2 -V 5 --vbr-new vs -V 5 --vbr-new -ms