I've read the recommended lame compile is 3.90.3. And that 3.93 and above is not really recommended due to not being heavily tested etc.

I've also read that 3.92 is basically the same functionality as 3.90.2 but compiled with a different compiler.

So is 3.90.3 more recent and better than 3.92? And why should this be? Surely it should have been called 3.92.1? TBH not much is said in the FAQ's regarding 3.92. I've been using it for ages and get transparent results with aps (I admit my ears are poor at differentiating though). Although I am yet to try with 3.90.3

I'm certainly not an expert in mp3 or here at ha, but I'll try to bring some light in here... so:



3.90.3 is the recommended version atm, the offical 3.93 release is said to be bad tuned concerning the presents. this issue should be gone in 3.93.1.
many ppl outside ha use this latest official version.



well, that the whole point in my view, the use versions and settings that give you alone the best results, as this is always in individual thing (all ears are different).
So try as much out as you like, but in the end find something that fits your personal needs. this goal is certainly not primarly limited due to the date or number of an compile.
Altough in the case of 3.90.3 you can rely on heavy testing by the ha community, that is why it's recommended. 3.93.1 or 3.92 may be better or worse, but are mostly conciderd to be quite good.

I believe that 3.92 is the official release of 3.90.2, and thus much the same. 3.90.2 was, prior to 3.90.3, the recommended version here on HA. So there should be VERY little difference, in terms of quality, between 3.92 and 3.90.3. To be on the safe side though you may as well pick up Dibrom's compile of 3.90.3. I hope this helps!

edit: forgot a comma

3.90.3 adds the -Z switch by default to --alt-preset standard and extreme. That's the only difference between 3.90.2 and 3.90.3. You'll probably be OK with 3.92, but you'll want to add that flag manually after the preset for optimum quality.

Why hasn't anyone bothered to test LAME 3.92 thoroughly?

Thanks for the information guys.

Now just to go and find out what the Z switch does.

Well, off the bat... it's not advised that you use it with --alt-preset standard - because it's already there in 3.90.3.

I think because the 3.90.2 compile had "caught on" around here completely by the time 3.92 came out, and so nobody really bothered. Now that Lame development has slowed down, it might not be a bad idea to test v3.92 more thoroughly and even add it to the "recommended" list if it passes snuff... especially considering it's the version I use.

Since several months I'm reading, that 3.93.1 isn't recommended because it's not so much tested. Let's test it !

why bother? There is no way it could be better than 3.90.2, and it is possibly worse. It isn't even the latest version. If anything, 3.93.1 should get the testing.

Because it's widely used (as far as I can tell). That's the only reason I can think of, but I don't think it's a bad reason. There are probably other things worth focusing on, though.

If someone is willing to test, first the -Z switch should be included in APS. Then the tester should encode most of the test samples (new link, anyone ?), with preset medium, standard, extreme, insane, several ABR and CBR bitrates, and standard -Y too. The ideal would be someone capable of ABXing the test samples to begin with.
Then a period of wider testing, encoding normal CDs should follow. Then, if nobody reports any problem, it could become the new recommended version.

Which is a great explanation of why it hasn't been tested, don't you think?

No one is saying it will be better - they're just hoping that it won't be worse!

Given the amount of work involved, it's no wonder no one has tried.


Now, if someone were to come forward with a new version that acheives better quality and/or lower bitrates, then there would be some serious interest.

Cheers,
David.

good point.

Isn't it quite a bit faster at encoding?

v3.92? I believe it's slower, and also produces larger files. Purely from a subjective perspective (no proofs, so this is just an empty claim) the files sound a little different than 3.90.2 as well... could just be my imagination.

As far as I can tell from the changelog there have been no changes in the psy-model between 3.90.X, 3.91 and 3.92. AFAIK the only significant changes are the support for --alt-presets at bitrates < 80 kbps (which are also included in John33's modified compile) and some changed compiler options in the make file regarding integer rounding, which caused the slight difference in bitrate.
Personally I'd assume that 3.92 is fairly save to use and should - quality-wise - be up to 3.90.3 when used with the -Z switch.

dev0

3.92 produces smaller files (5-10kbps) which should be of equal quality to 3.90.2 which is a good reason to do some ABX testing IMO.

My mistake on the size... I knew it was one or the other.

I don't really care enough to do ABX testing... enough people have said there's no reason for the files to sound different, and I'm not trying to prove anything. I added the appropriate disclaimer to my comment, so... yeah.

I'm a bit surprised that this question keeps coming up, despite the fact that it has been covered many, many times before.

3.90.3 is recommended by HA and 3.92 is not recommended by HA because after 3.90, there was very little to no attention paid by the LAME developers in making sure that the quality of the presets were not being degraded by changes that were made to LAME. Around the time 3.90 was released, I started to become more and more frustrated with the LAME developmental structure and eventually lost interest in trying to uphold the standard that I was eventually able to acheive at 3.90.2. Because of this, there was really nobody to routinely test and make sure things were not breaking or getting worse.

I do not believe that 3.92 and the MSVC compiles are of the same quality as the ICL compiles. I've already stated that at least on one occassion I have been able to discern a difference between files encoded with an ICL compile and those encoded with an MSVC compile. I don't have any in depth test data about the matter at this point because all of this occurred so long ago. I also do not have any plans to do this kind of testing on LAME again at any point in the forseeable future, so feel free to take this account with a grain of salt.

The fact of the matter though is that the alt-presets were tuned with a specific rounding behavior in the ICL compiler which is not the same by default in MSVC. Regardless of how significant this difference truly is, it will have an effect on the tuning that has been done.

I think the bottom line then is that if you want to make use of the alt-presets, you should use them in the context in which they were originally developed (ICL compile, 3.90.3). If you instead trust the non-alt-preset work and simply have faith in the fact that the changes to LAME which were made by the developers over the 3.9x versions didn't have an effect on quality (keeping in mind that nobody was really checking this thoroughly), then feel free to use 3.92 or any other non-HA endorsed version.

If it's any consolation, the difference should be small in any case, but if you're worried enough to be asking this question in the first place, the best bet is to simply use an HA endorsed version.

Dibrom -- I don't mean to be rude, but do you hold yourself to the same standards as everyone else here?

You seem to have just made a claim here that v3.90.2 and v3.92 (and/or other MSVC compiles) sound different, and you feel that quality is different. Please provide ABX results verifying this claim -- or correct me if I'm interpreting you incorrectly.

This is different than some postulation for purposes of casual discussion, like I made on another thread recently about WMA "sounding warm" -- you seem to actually be claiming a tangible difference subject to ABX verification. I'm curious if you're inclined to provide ABX results verifying this claim, if indeed that's what it is.

:-O

Am I wrong? I'm open to correction on this, and an apology if I misinterpreted what Dibrom was saying.

No, I'm just afraid of what's going to happen here. :B

If something negative happens, then afaic this board is no longer worth my time. So don't be too concerned about it -- life's too short.

I established the board here based on my own principles, so I believe this is a rather moot question.



I've spent enough time with LAME and I'm done with it at this point. I've spent countless hours testing and tuning LAME, and I have no plans to do any more of this publically, even to verify claims. The only listening tests regarding LAME that I am interested in at this point are of the sort that ff123 and roberto have worked on.

The only reason I've made this statement is so that people here are aware of the view held by the creator of the alt-presets and to point out a few issues regarding the whole quality bit. I could care less about proving a point, and so I'll leave it up to others (someone who hasn't put in all the time I already have) to do so if they like. I have absolutely no desire anymore to put in the extensive amount of work required (it would take many tests to come to any sort of meaningful conclusion to this whole question) to simply prove a point to you, especially when I've done so much in the past and nobody else has bothered to pick up where I left off or even to really continue (in any significant way) any of the work I started.



I'm not sure I'm claiming tangible differences as some sort of absolute, but I'm definitely claiming that a non-difference is certainly unlikely given my experience and the facts of the situation.

Heh.

I almost have to wonder if you consider the board worth your time even now given that you'd say something like this.

Given this and your apparent newfound subjectivist views, I sort of wonder what the deal is really..

I consider my views to be very much middle-of-the-road at this point. There is such a thing as feeling that neither the objectivist nor the subjectivist views accurately portray reality, and that's where I'm at right now.

Both of these are viewpoints, and are situational. Clearly, there are aspects of audio subjectivism that are totally foolish (like cables sounding different, outside of capacitance issues and such) -- but perhaps contrary to the views of many around here, I feel that there are objectivist views that are just as closed-minded. I don't see anything preventing a merging of the objectivist/subjectivist views (call it a 'higher synthesis' if you will) aside from a stubborn clinging to one's beliefs displayed by both sides in this issue, and an attitude of cynicism, disbelief and dislike on the part of both camps. Maybe I should mention an enjoyment of and an excuse to argue and debate, as well -- not a 'bad thing', and nothing wrong with it... keeps life interesting.

That said -- yes, I've been finding HA boring lately (not the first time I've said it). Why dissemble about this? I'm posting a lot less here lately as a matter of choice, and not necessarily because of some objective/subjective thing. There's just not much going on here recently that interests me personally. Que sera, sera and all that.

dude, read dib's post again. he said there are POTENTIALLY problems there, but none that he is aware of. There are no POTENTIAL gains to be made by using 3.92, so why go there?

Because it's there.

thread closed..