I asked this question in the following thread http://www.hydrogenaudio.org/forums/index....ic=26956&st=125, but realized it was a bit off topic and moved it here.

Does anyone have a definitive answer/guide at what bit rates stereo should be used vs. joint stereo? I have heard that, except with LAME, stereo should be used at higher bit rates for most encoders. With LAME it's easy "just use the pre-sets stupid and forget about it". However, with Fraunhofer, iTunes, Blade, Xing, and other MP3 encoders, it is not so clear.

Any knowledge on this subject?

Blade doesn't feature joint stereo.

A well-coded joint-stereo implementation will always yield a better perceptual analysis and compression than a forced stereo one. It's just a question of "figuring out" which frames need to use stereo because they have more than 50% of their sound signal different in one channel than from the other. The mid-side coding will be best for frames which have less stereo separation.

Figure it out this way. Let's say you have a true mono recording, and you want to encode it to MP3.

If you encode it in forced stereo, each frame will get half of your bits, whereas if the encoder uses mid-side encoding, since the side part is null (complete silence, compared to the mid signal), then you have all your bits for a single waveform, and a better possibility to restitute the signal as transparently as possible.

The problem in this choice of stereo vs joint-stereo (or so I've heard) is INTENSITY STEREO : it compresses more efficiently, but induces artifacts which can be audible when re-listening to the mp3 afterwards. (this whole little paragraph is interpretation and wild guessing, except for the part that says it's a problem )

So here's the tip,
stick to JS if you can.

Just to jump in on this discussion, if JS is "preferable" and Fraunhofer will only give you JS at bit rates 128 and below. Is there a way to attain JS above 128 with Fraunhofer.

Or is it , that since you are going higher in bit rate anyway >128 (closer to the original source), that JS isn't needed?

I guess the question, I really would like ask (forgive me, I did not do a search), is why did Fraunhofer limit JS to 128 and below? What is there reasoning for not allowing it(JS) for higher bit rates? Or did I answer my own question above?

@ezra2323,

I hasitated to convert my CD's to mp3's for years until I found a suitable encoder. It was a toss up between LAME v3.90.3 (--alt-preset standard %s %d) and Fraunhofer - Professional Codec (l3codecp.acm). Up until we discovered ,l3codecp.acm v3.3.0.44 with WMP 10 installed) I had a copy of Radium's, but read some negative reports about it on this forum. So I stalled on converting my CDs to mp3.

Now that I just purchased an mp3 (creative labs 1gb Muvo (running, gym)), I decided to convert my collection to mp3's using l3codecp.acm v3.3.0.44 at 128 bit/s). My reasoning, more songs at a lower bitrate, using the best encoder for such bitrate. Plus on the other thread you listed above, rjamorim indicates that Fraunhofer may be better tuned than LAME. I assume in general overall.

You should never assume.

@ezra2323:

Try LAME v3.96.1 with -V 5 --noreplaygain --athaa-sensitivity 1. See if that works for you.

Just waiting for someone to post such a reply. Rememeber the old Odd Couple TV series. Felix Unger "Never assume...it makes an A__ of you and me". Watching the old TV series I see.

Or from the mouth of Samuel L. Jackson:

" See you're assuming I won't shoot your sorry ass, and everyone knows when you make an assumption, you make an ass out of you and umption."

This thread has been quiet for a few years, but it showed up in my "search" so I am using it to post my question.

I've been making MP3's from CD's for about 9 years, and have been using Fraunhofer's MP3ENC31 Professional command-line encoder (v3.1, vintage 1998, cost me $199 as I recall) for all of that time (at 192kbps -> stereo). I've been using Audiograbber 1.83 to grab, specifying MP3ENC31 as the external encoder (with -br 192000 -qual 9). This has worked perfectly for me, and the MP3 quality is superb.

However (1) the performance of this old encoder is VERY VERY SLOW (as is well known), and (2) it only runs on Win98, being incompatible with the DOS Command Prompt window of WinXP. So I've been forced to retain a bootable copy of Win98 just for this program, for whenever I want to make some new MP3's from newly purchased CD's.

Recently I learned about the availability for free of the Fraunhofer Professional encoder, with the XP-incompatibility problem resolved. In fact, now it is distributed as L3CODECP.ACM (with Windows Media Player 10 and later), and all it takes is some Registry work to make it available as an "internal" audio codec. I found a web site where v3.3.2.44 was available, but the one distributed with Windows Media Player (at least with v11, which is what I now have installed) is v3.4.0.0.

Both of these L3CODECP.ACM files seem to work perfectly as "internal encoders" for both Audiograbber as well as Exact Audio Copy (EAC), and probably other rippers as well. And the fact that this version is usable under WinXP means that once I've convinced myself the audio quality of using this newest Fraunhofer encoder is at least as good as that produced by MP3ENC31 I will be able to ditch my Win98 environment forever. Also, L3CODECP.ACM supports 224kbps and 320kbps, something MP3ENC31 does not. Also, it appears that L3CODECP.ACM is dozens of times faster than MP3ENC31! And all of this is good.

So, to perform some experiments and comparison, I produced various audio files from the Sheila Nicholls track "Fallen For You" (from her "Brief Strop" CD), because it's all vocal and piano, and should easily show up distortion or ringing or other audio artifacts if the encoding is poor. I made 13 files, as follows, all ripped using EAC (v0.99pb3):

Original CD: WAV 34,640,300
APE 20,747,007 (lossless Monkey's Audio v4.01)

LAME v3.98b6: MP3 VBR 96/320kbps 6,321,261 (stereo) [quality="high quality", VBR quality = 0]
(average around 221kbps)

MP3ENC31: MP3 192kbps 4,714,357 (stereo)
MP3 256kbps 6,285,675 (stereo)

v3.3.2 ACM: MP3 192kbps 4,708,030 (stereo)
MP3 224kbps 5,498,880 (stereo)
MP3 256kbps 6,281,216 (stereo)
MP3 320kbps 7,850,034 (joint stereo)

v3.4.0 ACM: MP3 192kbps 4,708,352 (joint stereo)
MP3 224kbps 5,496,832 (joint stereo)
MP3 256kbps 6,279,168 (joint stereo)
MP3 320kbps 7,852,032 (joint stereo)

Note that "stereo" vs. "joint stereo" in any of the MP3 files was not my choice. It was simply forced by each of the encoders depending on the bitrate specified.

Also, I've been working very hard (using FooBar2000's ABX comparator function) to try and distinguish between the assorted MP3's and the WAV/APE and for the life of me I can only sort of recognize that the 192kbps results are very very slightly inferior to the higher bitrate results.

But I can't seem to distinguish between the "stereo" vs. "joint stereo" MP3's at identical bitrates (as produced by v3.3.2 vs. v3.4.0).

And I honestly cannot recognize any difference whatsoever between the LAME VBR result and any of the Fraunhofer results at 224kbps or higher, and only a small (but recognizable) improvement over the 192kbps files.

Now, this particular musical selection (only voice and piano) is, granted, not very complex. But piano levels and overtones in "Fallen For You" are dramatic, and yet all of the MP3 files seem to reproduce the WAV/APE results nearly perfectly, despite the obvious compression.

I'm still puzzled as to whether "joint stereo" for L3CODECP.ACM is the same as what Fraunhofer used to call "MS/IS stereo" in MP3ENC31, which is what was used for bitrates lower than 192kbps. "Stereo" was used for 192kbps and 256kbps.

And yet, it now appears with v3.3.2.44 of Fraunhofer Professional "joint stereo" was only used for the highest possible bitrate, 320kbps. And with v3.4.0.0 of Fraunhofer Professional "joint stereo" is now used for ALL bitrates! Does this mean "joint stereo" produces the BEST sounding results, when compared to "stereo"?

So, have they decided that this "bitrate sharing" approach between the two channels is ALWAYS GOOD, and is NEVER BAD?

Comments? Insights? I know... probably a subject that's been beaten to death. But these recently available encoders from Fraunhofer have gotten me interested in the subject again.

Obviously if I go forward with v3.4.0.0 (using EAC and WinXP from now on, never seeing Win98 again) I really have no choice, as no matter what bitrate I decide to use henceforth I will be encoding with "joint stereo" all the time. But I would like to hear a little editorializing from the group, if possible.

Thanks in advance.

While there is a theoretical advantage of joint stereo over L/R stereo, in the past there have been implementations that were flawed. Modern versions of any of the above mentioned encoders should be quite safe to use with joint stereo.

Note that while I refer to the advantage of joint stereo as being theoretical. actual examples of a difference in quality are very hard to come by. More likely what you will find is that the joint stereo version may be slightly smaller for equivalent quality.

Joint stereo can mean different things depending on context. Joint stereo can mean either intensity stereo or mid-side stereo.

To put it simply, intensity stereo is bad. Mid-side stereo is good.(*)

LAME's joint stereo mode uses a mix of mid-side stereo and left-right stereo, switching between the two as needed. If done properly, no quality loss is involved.(**)

(*Actually, intensity stereo is useful for low bitrate.)
(**For CBR, quality could actually be better, since mid-side encoding can "save" storage efficiency and create more "room".)

The logical and obvious choice is: Lame with Joint Stereo. End of discussion.

Back in 1998, the Fraunhofer-provided documentation for MP3ENC31 command-line encoder referred to it as "MS/IS-stereo". Looks like this was their term for what today is called "joint stereo".

Since they automatically reverted from "MS/IS-stereo" at low bitrates to "stereo" if you went to 192kbps or 256kbps, I always assumed that "stereo" was better than "MS/IS-stereo".



I'm learning.



But no intensity stereo (because it's bad)?



Doesn't apply to me and how I make MP3's. But I guess it would thus make sense for Fraunhofer to use it at low bitrates in MP3ENC31.



So you recommend LAME's joint stereo... when using LAME. And I assume I would also use VBR, rather than CBR. Best of all worlds, is the recommendation?

==> You would opt for LAME's joint stereo, over stereo? When placing LAME_ENC.DLL in the EAC or Audiograbber program directory, you can choose the LAME DLL as a "Waveform format" (for EAC) or an "internal encoder" (for Audiograbber). And then the DLL presents choices for stereo, joint stereo, dual channel, or force mono (in that order). Is this not an order that also implies "quality"? Is joint stereo genuinely superior to stereo?

Note also that the LAME DLL-approach does not support command-line parameters. But it has a separate LAME DLL tab for options: (a) stereo or joint stereo, (b) maximum VBR bitrate, and (b) VBR quality (0-high to 9-low). I used stereo, 320kbps max, and VBR quality 0, and was certainly very impressed with the results. I don't see a need to invoke the "external encoder" and specify command-line parameters unless there is a distinct reason or something I'm missing with the DLL-approach.


Back in 1998, I recall that the Fraunhofer MP3ENC31 was touted because it did not cut off at 16K but went up to 20K. Does LAME have any issues in this area? I have a recollection of trying lots of encoders out back in 1998 (including Xing) and while they all differed in speed what what was most important to me was comparing sound quality of the result (at 192kbps, where stereo kicked in for Fraunhofer). And as I recall it was MP3ENC31 which emerged victorious to my ears.

I've been using the Fraunhofer encoder (and 192kbps CBR, -qual 9, stereo) for so long that I'm of course hesitant to just change to LAME. And now that I've learned L3CODECP.ACM works under XP I was expecting to use it under EAC, with the same parameters specified (though not quite with the same command-line parameters). Making a change to LAME, 96/320 (or maybe 96/256), joint stereo, VBR and quality 0, would be a very significant mental paradigm change.

But, if that's the "best" approach, it's the result that counts... given that I'm using a lossy medium to begin with. The LAME VBR test MP3 file I made is about 33% larger than the 192kbps CBR result out of MP3ENC31. I'll also try LAME 96/256 and stereo and compare file size and sound quality, along with two more files made with "joint stereo" instead of "stereo" and then pick my "winner".

If you can get away with using a horrible old encoder like that, you'd probably be fine with 128k LAME/AAC/Ogg or anything else made in the last few years. I'd just use LAME -V 5 or maybe -V2 if you really don't care about disk space. No sense wasting much more time on this.

Let me guess. "Fallen For You" is 3 minutes 16 seconds, right?

The actual average bitrate for your LAME v3.98b6 file in this case is roughly 257kbps.

Yes, you should use lame's -m j switch (or join-stereo mode) with VBR, where it can save filesize. LAME can't do intensity stereo, so don't worry.

Regarding lowpassing (the cut off), not doing lowpassing is bad. Proper lowpasses are good.

edit: -m j

You can try..


as much as you like. but in the end, i bet that you will find LAME preset offered best quality (VBR+JS). I already wasted my time a bit to test these in the past, and LAME preset is best.

Just finished spending one hour listening to 1:48-2:18 of Olivia Newton-John's "Hopelessly Devoted To You" (1978) off of her 1982 "Greatest Hits Vol.2" CD. Granted, this was originally recorded three decades ago, and the CD itself is 25 years old so the engineering involved here is a bit "back-level". I would certainly not give this CD an award for sound production quality.

But this particular song, and in particular this 1:48-2:18 snippet, has a very prominent and very melodic bass line. The section is moderate in volume (not too low, not too loud) and the bass line is almost like a harmony track and "contra-puntal" to Olivia's voice. Wonderful balance.

Anyway, having listened to this track hundreds of times over the years, through my real sound system, top-notch Stax headphones, my Altec-Lansing 641 computer speakers, my Altec-Lansing 621 computer speakers, etc., etc., I am infinitely familiar with how it should sound... especially the bass line in this snippet.

So I compared the sound of (1) the pure WAV snippet vs. (2) the same snippet encoded with Fraunhofer Professional 3.4.0 (CBR at 192, 224, 256 and 320, -qual 9) as well as (3) Lame 3.98b5 (VBR stereo and joint stereo, -q 0 -V 0 -b 128 -B 320). Over and over, I listened to this :30 snippet, using Winamp 5.5 and the identical EQ settings as my player for all seven snippets.

The results? In my opinion, with the WAV snippet as the reference the Lame STEREO snippet was the MP3 winner. In my opinon, the clarity, definition, range and "punch" of the bass line was just about identical to that of the WAV snippet. I found the JOINT STEREO to be just a bit more "muddy" and less distinct, with a bit less "chest punch" as was clearly present in the WAV and Lame STEREO snippet.

What was interesting to me was my opinion that the Fraunhofer Professional CBR results seemed to have a small "de-emphasis" down low in the bass. While the 320 file certainly sounded superior to the 192 file, even the 320 snippet lacked the distinct clarity and "punch" of the WAV and Lame files (both stereo and joint stereo Lame sounded somewhat superior to even the 320 Fraunhofer file). I don't know if the Fraunhofer encoder has a low-pass filter of some kind active (I've never noticed that before, but then I've never done these kinds of comparisons before) but for some reason there is just a small de-emphasis down low that is obvious.

Finally, comparing the file sizes reveals the following:

WAV 33,092,684

FHG MP3 192 4,499,456 (joint stereo)
FHG MP3 224 5,253,120 (joint stereo)
FHG MP3 256 6,000,640 (joint stereo)
FHG MP3 320 7,501,824 (joint stereo)

Lame MP3 128/320 5,542,848 (joint stereo)
Lame MP3 128/320 5,646,336 (stereo)


So, all things considered, I've decided to go with Lame STEREO as my new WinXP-compatible encoder and setting going forward. It looks like my new MP3 files will be about 25% larger than my old Fraunhofer-produced MP3 files (using super-slow MP3ENC31 at 192 -qual 9 under Win98), but I'm not really concerned about that. It's the sound that matters, and I can tell that Lame STEREO is better, and that makes me satisfied.

Lame JOINT STEREO is good (better than Fraunhofer, I now agree). But in my opinion STEREO is even better than JOINT STEREO. Just my 2 cents.

My work here is done.

You really need to do your testing with your expectations removed from the picture. Familiarise yourself with ABX testing.

Foobar2000 is a great media player as it comes bundled with many tools, an ABX comparitor being one of them.

It is also a requirement of these forums when making qualitative comparisons to show your ABX results to back up your claims.

Good luck.

No, you cannot, because you haven't done any ABX-tests. See ha.org Terms Of Service #8.

Can you re-try the lame test with just a preset and nothing else. Why not just try "lame -V2" for your joint stereo test (then -V1 and -V0 if you think you can distingish V2)


Edit: Oh and yes please do an ABX test, you totally cant trust any conclusions otherwise.

This should be high-pass filter, i.e. one that attenuates frequencies below a cutoff frequency.

I believe that DSperber is correct in calling this low-pass, since low-pass means that only frequencies lower than specified value will be preserved (in other words, only lower frequencies pass).

Regards,
Irakli

"Lack of bass" doesn't sound like low-passing to me.

Actually, it sounds more like uninformed first impressions.

That's right, but DSperber is claiming the corner is "down low in the bass". If it were a low-pass then there wouldn't be much music left to listen to, correct?

Ok, ok... so I'm "uninformed". I'm obviously using phraseology that isn't precise or accurate. My mistake.

But what I was trying to describe was simply an effect where it appeared the low frequencies were "de-emphasized", i.e. "rolled off" below a certain low frequency point. In the MP3 file frequencies above that point seemed normal, but frequencies below that point seemed "less loud" than the contrasting sound from the original WAV file.

Maybe I should have called it a "low frequency filter", or better a "low frequency roll-off". Anyway, hopefully I've now described what I heard more accurately... no matter what we call it.


As far as using Foobar200's ABX comparator technology, I DID learn about it both from this forum and elsewhere on the net. And I DID download and install it several days ago but at first couldn't quite figure out exactly what I was supposed to do. Note that there isn't a HELP with the program, nor a PDF or other documentation. I guess I'm just supposed to intuit how it works from the interface dialog itself, and others' comments on forums and web sites.

And I apologize for not reading the "rules for joining this forum" closely enough to know that ABX output was required in order to support claims of "superiority", rather than just writing about subjective observations. This makes sense.

Anyway, I now understand how the ABX interface is used although I'm still practicing and training. It turns out it's REAL EASY to distinguish two clips that are obviously quite different in their quality, but is VERY HARD to distinguish two clips that are very VERY similar in their quality. The trick, obviously, is to find a very short interval somewhere in the track that is EASIER to distinguish between the two versions.

So I'm still working on finding exactly such a 1-3 second clip, rather than the 30-second clip I'd mentioned earlier. And then I will do the ABX and see if my "uninformed impression" that STEREO seems superior to me than JOINT STEREO (using the Lame -V 0 -q 0 -b 128 -B 320 other highest-quality settings) is supported or contradicted. Hopefully, there will be somewhere in some track such a "dead giveaway" short clip where there is a clear difference in how these two methods perform, with one or the other being the "winner" (though the margin of victory may be small).

Don't take the challenge to what you hear, or think you hear, personally. The superiority of joint stereo over stereo, while theoretically justified, is very difficult to prove or disprove in practice. If you are in fact hearing a difference then we all want to know about it, but without ABX results nothing will be accepted.

By the way, what ABX testing does is verify which thing sounds more like the original, but it is still possible that the one that sounds more like the original does not sound as good TO YOU as the other. In the case of high bitrate lossy encoding where all codecs sound pretty good, what counts is accuracy. The goal is something that is indistinguishable from the original. At lower bitrates where no lossy codec sounds exactly like the original, a different kind of evaluation is used that is much more subjective.

Well please answer two questions I have about this ABX process as it particularly applies to the question at hand, namely (1) what I am supposed to compare as A/B in how many different tests, and (2) how I am supposed to use the results of the multiple tests to conclude that one or the other MP3 "sounds more like the original"?

In other words, I actually have THREE files of interest: WAV original, MP3 from JS encoding, and MP3 from S encoding.

So, what would be A and what would be B, in what separately constructed tests, in order to answer the questions: (1) which sounds more like the original, and (2) which sounds better to me? I guess question (2) is easy, by just comparing the two MP3 files and deciding which I prefer, and if about 100% of the time it turns out I always pick S or JS as the one which "sounds better" then I guess that's the answer to that question. But what about question (1)?

It seems that the ABX test allows me to determine if I can distinguish between two versions (either simply to identify them or to prefer one over the other), and how reliable is that distinction (about 100% indicating a clearly recognizable distinction). Assuming I compare either MP3 to the WAV and can always pick which is which because I've found a test clip that easily reveals which is "better", which likely points to the simple fact that the WAV is going to be "better" than MP3 (not surprising), that still doesn't tell me which of the two MP3 files sounds MORE like the original WAV, unless I can compare all three simultaneously.

Preference is a subjective thing.
But as your hypothesis is of the kind 'the stereo mp3 version sounds like the original, the joint stereo version doesn't' this would be backed up if you can abx the joint stereo version against the original but can't do it with the stereo version though you try as hard as with the joint stereo version.
In case this your result your listeing test proves your hypothesis for you as the listener. If that's correct for you as the listener it is supposed to be correct for other listeners too, but confirmation is required by other listeners. So it would be nice if you can share your problem snippet.

For having a valid abx result it is necessary to do for instance 10 trials (8 is absolute minimum), and you should be able to guess correctly nearly each of the trials. Watch the guessing probability shown by foobar.

For the actual mechanism of performing ABX I suggest you start by testing the JS mp3 against the original wav file.

Here are the recommended steps.

1. Start with just the wav file and load it into a playlist in Foobar 2000

2. Use foobar the convert the wav file to mp3. I suggest you just start with using "lame -V2" and no other parameters for the making JS file. It reduces the possiblity of hitting a bug when using some less well tested parameters. BTW : To convert the wav to mp3 in foobar just right_hand_click the wav file in the foobar playlist and select "convert" and follow the onscreen instruction from there.

3. Load the newly created mp3 into the foobar playlist so that it now contains just two tracks, the original wav and the mp3.

4. Highlight both tracks (as usual, hold the "ctrl" key while clicking to select tracks to select more than one item) then right_hand_click on the selected tracks and select "utilities"->"abx two tracks".

5. Now it's just a matter of matching up which of X or Y is A. Remember to be objective you should decide before-hand on how many trials you are going to do (10 is recommened) and not just "quit while you're ahead" as that also introduces bias into the test.

Yet the discussion continued merrily on despite your edict.

Just wondering if you have any updates on the Lame -V2 versus wav ABX DSperber?

Over a week and no results so it's starting to look like another case of audible differences that suddenly disappear once an ABX test is introduced.

Not casting any aspersions on the original poster though because I think most of us have underestimated the power of our expectations in subconsciously colouring what we "hear". Before learning about ABX testing, I know I did.

Sorry for the delay. Just got back from a holiday out-of-town since the beginning of December.

I'm absolutely still aiming to pursue the ABX project (and post results here) to see if I can actually and definitively tell the difference between the various encoding options I have before me. Key to this, of course, is being able to luck into a short snippet that has clearly distinguishable differences based on encoder method and parameters.

I really do need to make a decision, because I have a stack of new CD's which are due for "MP3 production" and I want to get going.

Haven't vanished.