How to get rid of joint stereo, lame is forcing joint stereo on every single file Options

[paappo]

Because all of this techo-mumbo-jumbo goes over my head, I want to try on my own what sounds best. I dont think Im "enlightened one".. Im sorry, but I dont blindly believe what I read on the web (maybe its a generation thing) I should have added that to the first post :I

[odyssey]

Because all of this techo-mumbo-jumbo goes over my head, I want to try on my own what sounds best. I dont think Im "enlightened one".. Im sorry, but I dont blindly believe what I read on the web (maybe its a generation thing) I should have added that to the first post :I

Exactly, but I think you would want to refine your way to determine the best sounding one, as your way is flawed. Just encoding the mp3's to a CD and knowing what's playing, your mind can be biased towards the one you *think* sounds best. This is called placebo.

Instead I think you should get a good set of headphones and connect them to your onboard sound. I don't believe it's as bad as you make it (maybe more likely the speakers connected) or maybe if you have an SPDIF output you can connect it to a receiver. With such setup you can use a program like foobar2000 to do an ABX test. This way you can easily find if you are able to hear a difference from two different encodings.

Edit: However, why do you think "real stereo"-mode would sound better than anything choosen to be the default (and therefore preferred)? Here on HA, many has conducted (ABX) tests of especially problematic samples to improve the lame algorithm in any possible way.

I think that if you have mistrust to a lossy encoder, maybe you would be better off with a lossless one, like FLAC.

This post has been edited by odyssey: Apr 15 2009, 10:39

[paappo]

abx eh? I'll look into that. Sounds easier than burning clips to disc.

I have this image in my head, that if you have an 'raw' audio track and you somehow mix it or join frequencies, it won't be as 'pure' as the original. And if you encode both tracks separately, they would sound better. But I suppose this is herecy and you already explained it but like I said; I dont get all of this techno-mumbo-jumbo, I just want it to sound good in my ear, I dont care if you mock me for it

Yea but flac takes more space on a disc (i think) and i dont even know if my player supports it.

[LaserSokrates]

It's always the same here. The same questions always spawn the longest and most pointless discussions. Top favorites: "Why is joint stereo default when stereo is better" (coming from people still using LAME 3.92 with FULL STEREO) and "Is lossless really lossless" or, in it's variation "I clearly hear a difference between the FLAC file and the WAV" (even though they never ABXed). I wish these threads would get one answer only and then be done.

[odyssey]

I have this image in my head, that if you have an 'raw' audio track and you somehow mix it or join frequencies, it won't be as 'pure' as the original. And if you encode both tracks separately, they would sound better.

Not nessesarily. mp3 is a lossy compression algorithm which uses many techniques to save space. it's optimized in a way that anything that humans can't hear are filtered out. As a result, most encodings around 128kbit (something like -v4 or -v5 I think) would be completely transparant. Joint Stereo is one of the techniques to save bits, but if you care about pure, then Joint Stereo should be the last worry!

Yea but flac takes more space on a disc (i think) and i dont even know if my player supports it.

FLAC is something like 5-7x the space of properly encoded mp3's, but to me it sounds like it would be a good compromise for you (granted, your hardware supports it - much hardware already does!).

[Mark7]

I have this image in my head, that if you have an 'raw' audio track and you somehow mix it or join frequencies, it won't be as 'pure' as the original. And if you encode both tracks separately, they would sound better.

Well, midside is not so hard to understand. Here are the formulas to get midside (and back to stereo) explained with some examples:

The formula to get midside:
(L+R)/2=M
(L-R/2)=S

Let's take these values for Left and Right:
L=4
R=3

So midside will be:
(4+3)/2=3.5
(4-3)/2=0.5

And now back from midside to stereo we need this formula:
M+S=L
M-S=R

With the numbers:
3.5+0.5=4
3.5-0.5=3

Now an example when L and R are the same (mono music):
L=5
R=5

So midside will be:
M=5
S=0

Now you see that the side channel is 0. That is great because this means that no bits are needed for encoding. All of the music is present in the mid channel and since you don't need to use bits for the side channel, so you can use twice the bits for encoding mono music.

When L and R are too different, it doesn't make sense to use midside. So in that case lame will use normal stereo.
Lame will determine for each frame (very small piece of music) if midside or normal stereo is more efficient.

[paappo]

Oh my god.. what part of my reply did you not understand "I have this image in my head..and you already explained it but like I said; I dont get all of this techno-mumbo-jumbo" Anyways.. I fired up that abx test and.. oh the humiliation! Atleast from the first test I could not tell which one was joint stereo or forced stereo. I'm really confused here, I'll do some more testing.

Thank you. You really saved me a lot of trouble with that abx thingie
Next time I WILL read the f.a.q. and use search. Now slap me and close the door.
Apologies for poking the hive and upsetting everyone, that was not my intention. Now you can have your laughs and point fingers. ok baibai.

@lasersokrates
Sorry man.. You have to just face the fact that the web is full of idiotmonkeys like me who ask stupid questions

[odyssey]

No need to slap anybody We all started somewhere.

Try to make a couple of encodings with different (lower) bitrates. At some point you will find that the full-stereo encoding will suffer in contrary to the joint-stereo. But just for fun

[sld]

It's quite sad when people who want music to 'just sound good' are willing to stumble over semantics. 'Joint-Stereo' is a misnomer, and if they had taken just 10 mins or so reading the resources on this forum, they will have saved themselves a lot of time and effort better spent on enjoying good music.

Once you find a good information lode (like HA.org), all the good work has been done by users and developers before, leaving us to share in the fruits of their labour.

This post has been edited by sld: Apr 15 2009, 12:54

[uart]

Yep I took a look at the usual filesharing networks recently (just for research purposes ). I was amazed to see how many files there are still encoded with CBR and L/R stereo, and this was true even of files that were made with reasonably recent lame versions.

I have this image in my head, that if you have an 'raw' audio track and you somehow mix it or join frequencies

Joint stereo does not mix or join frequencies in any way, what made you think that it did this?

If you want a simple non-techno-mumbo-jumbo view of what the "joint" means in JS then think of it as simply allowing the L and R channels to be analysed together (rather than forcing them to be treated totally separately) so that better compression can be achieved by not having to encode two copies of the same data when the left and right channels contain similar audio information. Compression is the goal of mp3 encoding, better compression means that we can achieve the same quality with less bits or alternatively we can get higher quality with the same amount of bits. (yes this is an oversimplification of JS but I think it's appropriate here).

This post has been edited by uart: Apr 15 2009, 13:55

[odyssey]

There are many myths around the world that people believe. It's hard to get rid of a myth once established no matter how much a single source (i.e. HA) tries to convince people that they have the wrong impression especially when many other sources are spreading misinformation.

[2Bdecided]

Until the bandwidth of the components is limited, there is no difference between component and RGB as the transcoding is lossless. It's just a lot easier to manipulate levels and saturation in component than RGB

In digital video using any of the standard formulae, the transcoding certainly isn't lossless, as there is re-quantisation involved.

With typical 8-bit data, this re-quantisation is easily visible on smooth gradients. The problem can be reduced by changing the formulae to produce non-standard video (e.g. 0-255 for Y rather than 16-235), but it's certainly not lossless.

Cheers,
David.

[kiit]

Joint Stereo is LOSSLESS. (Except in some very old encoders that had other very bad issues anyway as someone already noted.)

Lossless == Lossless. It isn't some analog process, its like saying 2 + 2 == 4. It is only with analog processes you get variations.

This has been explained here in-depth a lot, I am just trying to be concise and not use technical-jargon.

MP3s however, are lossy, but the Joint Stereo part of LAME is completely lossless. Not using Joint Stereo with LAME is just taking away bandwidth from the lossy part, making them sound worse, whether you notice that or not.

(Hope I said that right, *shrugs*).

(Edit: For best possible sound (mathmatically speaking), use a completely lossless codec like FLAC or Wavpack. Most people cannot tell the difference between good LAME encoded lossy MP3s and Lossless FLACs, but lossless has other things going for it.)

This post has been edited by kiit: Apr 15 2009, 16:06

[pdq]

Just to add to the mid/side encoding is lossless argument, even lossless encoders use mid/side encoding if it saves space. If mid/side encoding weren't lossless then you couldn't use it in lossless encoders.

[timcupery]

Joint Stereo is LOSSLESS. (Except in some very old encoders that had other very bad issues anyway as someone already noted.)

Lossless == Lossless. It isn't some analog process, its like saying 2 + 2 == 4. It is only with analog processes you get variations.

This has been explained here in-depth a lot, I am just trying to be concise and not use technical-jargon.

MP3s however, are lossy, but the Joint Stereo part of LAME is completely lossless. Not using Joint Stereo with LAME is just taking away bandwidth from the lossy part, making them sound worse, whether you notice that or not.

This is true, and seems to be missed by some of the people in this thread, not just the OP. Joint Stereo, as used by Lame and all other modern quality mp3 encoders, is a process that simply allows more quality to be packed into a given bitrate, or the same quality to be packed into a lower bitrate, because it can represent the exact same audio stream with fewer bits.

As pdq noted in the post directly above mine, even Lossless algorithms (which reproduce exact data as an original wav file) use Joint Stereo. It's one of the major ways in which they save space over a wav file.

The main factor for how much space saved by Joint Stereo, is how similar are the L and R channels? If they're very similar, then the two channels can be represented by only slightly more data that it would take to represent a mono channel at the same quality level. If the L and R channel are very different, Joint Stereo saves almost nothing and the file for a given quality level will be close to twice as large as a mono file at the same quality level.

I started a thread on this a few weeks back, did some testing (on one song, with about average difference between L and R channels for normal pop/rock music) and found that (for the file I tested), 128 kbps joint-stereo was about the equivalent quality to 160 kbps forced stereo. All other things being equal.
Here's the thread:
http://www.hydrogenaudio.org/forums/index....showtopic=70722

This post has been edited by timcupery: Apr 15 2009, 20:58

[kornchild2002]

Sorry man.. You have to just face the fact that the web is full of idiotmonkeys like me who ask stupid questions

I don't think that you are an idiotmonkey, just someone who was either using old information or going off of false information from anti-Joint Stereo people. We all start off somewhere and hydrogenaudio is a place where people can learn lots of information. It is just that there are some common thread topics that have been discussed ad nauseum. The "issues" with Joint Stereo, accusations that lossless really isn't lossless, accusations that Apple lossless really isn't lossless, the "benefits" of PCM WAV audio versus lossy and lossless compressions, and so on. Many people who make false posts regarding those topics are often thick headed self-proclaimed audiophiles who aren't willing to budge. They come here mainly to stir up the pot. It is often hard to differentiate between these audiophools and people who are actually here to learn. It sometimes takes 3-4 posts from these people before we can actually tell the difference.

So don't take the negative comments in vein. Some people (including myself) just get tired of seeing the same old arguments over and over again with many of these topic starters being nothing more than thickheaded audiophools who want nothing more than to stir things up.

[2Bdecided]

You people shouting "JOINT STEREO IS LOSSLESS" are desperately oversimplifying the issue. It's all been discussed before, and I won't repeat it for fear of confusing the original poster - but I don't think stating a misleading oversimplification as fact helps anyone.

Cheers,
David.

[Surdaster]

In what cases does joint stereo introduce losses? I don't think the original poster is here anymore, no need to worry about confusing him.

This post has been edited by greynol: Apr 16 2009, 17:06

[pdq]

Interconversion between R/L and M/S is lossless other than rounding. That's not the problem. The problem is that that is followed by a lossy conversion, and there can be significant differences in how the lossy conversion is done in the two cases.

Psychoacoustic modeling makes assumptions about what will or will not be audible, but the modeling is not perfect. In particular, a model that works well in L/R mode for sounds coming exclusively from the left or right speaker may work less well for sounds coming from the center. In M/S mode, sounds in the center may be modeled well, but sounds on the left or right not so well.

The bottom line is that one or the other mode may be modeled better at any particular time, and we hope that the encoder has made the best choice, but nothing is perfect.

[user]

Thanks 2Bdecided to point out the apparently new myth (which does not get more true in repeating it) about joint stereo being Lossless in lossy encoders.
As he said, that is an oversimplification.
The term Lossless for describing joint stereo in mp3/lame or mpc cannot be used.

Short Examples, that people keep it in mind:

the (old) switches --ns-safejoint as switch preventing joint stereo in lame to be too agressive, being on the safe side.
the general and variable switch ms-fix , which enabled customer-controlled values for lame, to switch between stereo and joint-stereo mode inside the "joint stereo mode", similar to ns-safejoint.)

ms-fix was or is obviously an experimental switch.
I recall somebody testing a lot at old times, who found a certain ms-fix value for his hearing, to be safe for the joint-stereo-mode.
(Also the ns-safejoint).

If joint-stereo would have been Lossless, there would have been no necessity for ns-safejoint or ms-fix switches.

Today, the result of such experimetal switches will be or should be built simply into the recommnded standard lame switches, like -V5 , -V0 , -V2 whatever.


Another example from another lossy encoder mpc:

MPC used joint-stereo, too, to be efficient.
The higher the quality target, variable bitrate target, q-setting, the higher the built-in "ms switch".
Eg. -ms 10 , ms 12 or --ms 15 (being --ms 15 standard for --quality 10 , a high quality setting over or around 300 kbit/s.)


For modern encoders i recommend simply to use the recommended standard settings,
like -V settings for mp3/Lame, --quality x for mpc , eg. mpc --quality 7-8 --ms 15 to be at q7 or q8 quality/bitrates with the safety of q10 stereo-image --ms15,
or going Lossless FLAC -8 -V (-V = verify output) , flac as true Lossless format being independent of sound-quality used on switches, same original quality, if used default -5 or -8 , -8 which needs simply more cpu power/encoding time to compress on the win for few extra MB space.

This post has been edited by user: Apr 16 2009, 21:51

[2Bdecided]

There you go Surdaster: pdq and user have covered it.


btw there are situations where forcing discrete stereo can cause problems - it forces the coding noise to be decorrelated between the two channels. Now if the signal itself is highly correlated between both channels, it would be better if the coding noise was too. Someone has already mentioned the binaural masking level difference, and that's exactly the issue.

Of course you can bump up the bitrate so that the coding noise, correlated or decorrelated, is so low as to be inaudible - but the masking difference between correlated and uncorrelated can be (I think) over 10dB - do you think mp3 always has the bits to spare to overcode everything by 10dB?


Now, in practice, it can normally do well enough in these circumstances - 160kbps per channel isn't too bad. Whereas the situations where the joint stereo algorithm goes wrong (if it does) can be clearly audible (though AFAIK they've all been fixed long ago).


So it's not all clear cut - but on balance adaptive joint stereo is the way to go - clearly more efficient, and potentially better quality. Anyone who claims that discrete stereo provides better quality overall is wrong. Anyway who finds an isolated problem sample should bring it here!

Cheers,
David.

[timcupery]

Thanks to 2Bdecided, user and pdq for setting me right on this.
I was thinking about this more after writing my post, and thinking of some of the reasons why lossy encoding could mess with the losslessness of basic m/s algorithm, and you guys laid out a few more as well.

[Surdaster]

Thanks, all ... Would it be safe to say that joint stereo is a lossless process, but how any given lossy encoder (or psychoacoustic model) handles the output of that process is variable? If so, are there any other lossless processes that can confuse a lossy encoder and produce negative results?

Thanks for indulging my idle curiosity.

[odyssey]

If so, are there any other lossless processes that can confuse a lossy encoder and produce negative results?

Does that even make sense? You don't need to worry about the internal procedures of how an mp3 is encoded at all.

Maybe your concern can be defined as "Will an mp3 encoder eventually produce audible artifacts?". In that case, yes it can, but the algorithms are optimized (especially in lame) to produce less artifacts even on specific problematic samples. Although this are usually only noticable with ABX tests and not regular listening.

If you dislike the fact that mp3 is a lossy process that are different from the source, go use a lossless format instead! IMHO mp3 are suitable for anything that requires low bandwith or small space, e.g. music on my laptop, dap or streaming services. For regular listening I prefer my lossless collection stored on a large NAS.

[pdq]

Thanks, all ... Would it be safe to say that joint stereo is a lossless process, but how any given lossy encoder (or psychoacoustic model) handles the output of that process is variable? If so, are there any other lossless processes that can confuse a lossy encoder and produce negative results?

Thanks for indulging my idle curiosity.

I think you could say that surround-sound encoding falls into that category.

Lossy stereo encoders make assuptions about what is audible and what is not. Surround-sound encoding results in things being audible in a surround-sound system that would not be audible if played back in a stereo system.

The result is that, typically, forced-stereo encoding sounds better for surround-sound than joint stereo. This is entirely the fault of the assumption, inherent in the encoder's psy model, that the material will be played back in stereo.