Hi all,

I'm just wondering. All this talk about earlier versions of LAME and --alt-preset-standard is really confusing to anyone visiting this site. Those threads are getting realy long now and they don't seem to answer all the questions needed to be asked.

Are the LAME versions improving or are they not. Dibroms 3.90.3 compile is quite dated now and quite a few revisions to LAME have been done. LAME is being compared to Itunes now using later compiles of LAME, does this suggest improvement?

Why can't we have a secured thread that only can be updated by an Admin or something like that to explain what version of LAME should be used with "these" settings. That way a newbie like myself would wander in and say, oh, ok, I should use LAME version xxxx with the setting --preset standard etc, not wade through 30 odd pages trying to find out what's going to give him/her the best results.

Dibroms 3.90.3 compile seems to be the only compile worth using for encoding music is this True/False?

A lot of people seem to think that only Dibroms 3.90.3 compile is worth using. With all the development going into LAME, is this the case!, is LAME improving or in fact giving worse results from when 3.90.3 compile was done.??

EAC the CD ripping program, seem's to pull default LAME settings from somewhere on the WWW. Is the author aware of the setting changes and revision changes? This can affect quality. Just wondering where it get's it's info from. is it this forum/website?

Lame 3.96 preset standard using less than 128kpbs frames. I'm not sure if it effects the quality maybe someone elles can tell us,



LAME 3.90.3,

That's all fine, but it still doesn't answer my question properly. 3.90.3 Is using q2 where as the 3.96 is using q3 which is lower in quality, however, is it just that suddenly 3.96 is able to achive similar quality sounding file because it is more efficient?

What a newbie want's is not to be flooded with pages of "Hey look at my encoding of file xxx.xxx " and then 65 other peoples encodings of x, y, z and others, but to have a recomendation of what would now be the recomended LAME revision to use.

I also would like to know, is LAME improving! is it now able to encode more thouroughly, is it more efficent, have the changes to the code had a significant impact on quality or is 3.90.3 the best LAME can offer.

[quote=westgroveg,Jul 20 2004, 01:42 AM]
Lame 3.96 preset standard using less than 128kpbs frames. I'm not sure if it effects the quality maybe someone elles can tell us,



That's all fine, but it still doesn't answer my question properly. 3.90.3 Is using q2 where as the 3.96 is using q3 which is lower in quality, however, is it just that suddenly 3.96 is able to achive similar quality sounding file because it is more efficient?

What a newbie want's is not to be flooded with pages of "Hey look at my encoding of file xxx.xxx " and then 65 other peoples encodings of x, y, z and others, but to have a recomendation of what would now be the recomended LAME revision to use.

I also would like to know, is LAME improving! is it now able to encode more thouroughly, is it more efficent, have the changes to the code had a significant impact on quality or is 3.90.3 the best LAME can offer.

Why don't you take a listen and see for yourself?

If you hear improvements, then you'll know it's getting better.
If you don't hear any regressions, then you'll at least know it's not getting worse.

The List of recommended LAME settings gives recommendations regrading what settings to use.


Latest versions are improving; see the changelog for full list of improvements/bug fixes. They are being compared to other encoders because extensive testing has not been done, and comparisons help gauge performance compared to previous versions.


That's pretty much what the list of recommended settings is. It's only updated by the people in charge, after enough testing has been done to recommend different settings.


I wouldn't say it's the only compile. It's your safest bet for having the least chances of artifacts in your music, but later compiles are ok, too, and you won't notice much of a difference unless you run into problem samples (where later compiles could perform better or worse).


Both. In some cases it gives worse results than before, in some cases better. The goal is to improve it until overall it gives better in most cases.


EAC uses --alt-preset standard if High Quality is selected, and --alt-preset 128 if Low Quality is selected.

In conclusion, --alt-preset standard is the recommended setting. If you really want to make sure you minimize your chances of hearing artifacts, 3.90.3 is the recommend compile. If you want to use latest version despite not being tested as well, 3.96.1 is recommended. Who knows, you might find a bug or problem sample and be able to report it and contribute to development

I do, all the time. The problem is, I want maximum quality, compatibilty and minimal storage space. Hence I use .mp3. I am yet to find an encoder that I can't distinguish differences between the original source material (wave file) and the .mp3.

The all just have artifacts, the 3.90.3 has ringing in high symbol music like rock, and 3.96 too, mind you I thought it did a slightly better job. Others seem to think differently and say they can hear ringing in 3.96 not in 3.90.3. I produce and write my own music being in a rock band. Quality and transportation is important to me.
I came hear looking for answers because I have my onw conclusions and I wish to hear the thoughts on people who are experts in the .mp3 field and see what their findings were.

If you think that I haven't listed for myself, why would you think I would be using .mp3 in the first place or posted for comment?
I wanted opinions and for the most part it took 2 hours to go through the sticky threads that didn't offer any more insite.

I guess I will wait before upgrading to 3.96 or untill some proof that the "artifacts" currently being found, are reduced.

In conclusion, --alt-preset standard is the recommended setting. If you really want to make sure you minimize your chances of hearing artifacts, 3.90.3 is the recommend compile. If you want to use latest version despite not being tested as well, 3.96.1 is recommended. Who knows, you might find a bug or problem sample and be able to report it and contribute to development

[/quote]


I guess more listening tests need to be done and it's really up to the end user. We may all end up with our own customizable versions of.mp3 encoders in the end. Every ear seems to hear differently.
I would have thought mine would have been some of the worse as I play in a rock band. lol
Bring on the testing! we want quality!

That is not a problem. In your histogram sample, it seems to be silence. As you know, Lame uses the lowest possible bitrate when encountering silence. A frame is 2 granules with 2 channels per granule (if stereo). If those 4 parts are silent, then 32kbps will be used. If only some of those parts are silent, then only the silent part are reduced to 0.
That is how you can encounter some 40 or 48kbps frames.

I think that there is no big quality difference when using preset standard with 3.90.3 or 3.96.
If there was a big difference, it would be well known now.
Regarding lower vbr settings, I am confident that 3.96 is superior.

Personally I have always used 3.90.3. Seems to produce grreat results at aps and apx. Also, if you want small and good quality then ogg is probably better, though it takes a slight hit in compatability. Of course every software audio player that is worth using has ogg support by now I think. Plenty of hardware does to. But I still use mostly mp3 because it's the only thing my car's cd player supports. So it's really seems like 3.90.3 and 3.96.1 are a matter of preference for now unless you doing 128kbps.

...unless what? I know exactly what the original poster means, I've been doing the same thing for days, weeks even, searching through all these threads (some dating back to 2001) and it's 3.90.3 this and 3.90.3 that - I'm a IT bod myself and the idea of older software being better is an anathema to me.

But if it really is the case that 3.90.3 is still the one to use then I think those stickies should be updated to mention the latest versions, cos to a visitor it just looks like the threads are old and the information we need is missing.

I've been using 3.96.1 on quite low settings (abr 128 - full stereo (I know, I've just today found the info about j-stereo being better quality *slaps forehead*)) and it's not bad at all - slightly squelchy on heavy cymbals but that's about it.

Is LAME v3.96.1 simply the same thing as v3.96 stable ? And if not, where can I obtain v3.96.1 ?

Regards,
Madrigal

Nobody has ever stated that 3.96 is bad. Testing has even shown that it performs better than 3.90.3 (using --alt-preset standard) on some samples, but also worse on others. However there is not enough statistical evidence to replace 3.90.3 as the HA.org 'recommended' version.

I'm thinking of adding a note to the 'recommended settings' thread pointing to the new VBR presets (-V n) and recommending 3.96 for low bitrate VBR, but I'm afraid that this would only cause further confusion.

For CBR/ABR 3.90.3 also performs better on a lot of samples (look at the 3.90.3 vs. 3.96 test results).

i agree and trust gabriel, and all the LAME devs ... 3.96 rules!

I think it might be time to switch to 3.96 as the recommended version, even though it's not as well tested as 3.90.3 and we don't know if it's better or worse at high bitrates.

1) It's definately better at low bitrates VBR. It would be nice to recommend it for that range, but having multiple recommended versions is confusing.
2) There's no evidence to suggest that the regressions at higher bitrates are more numerous or serious than the improvements. Several people have tried to test this, and have not been able to find a definitive answer, which tends to make me believe there isn't a huge difference between the two.
3) More current versions are easier to obtain for more people from more locations (e.g. if you have an obscure platform).
4) 3.96 is a lot faster.
5) It would get used more. If it turns out that there are serious regressions we can switch the recommendation back and/or fix the regressions.
6) 3.96 has some nice new features. The new -V settings that are synonymous with the presets make it harder for a novice to choose a bad setting.
7) It seems like 3.96 has shed some cruft, which is good, as a rule.

Please take a look at the test results, phong:
http://www.hydrogenaudio.org/forums/index....showtopic=20715

I've read that thread, and I agree with Phong. A little better on some test-cases, a little worse on others (with --preset standard) but with lower bitrates, faster encodes, and much better low-bitrate quality (especially for VBR!)

The discussion seems to have stagnated since then... someone make a decision!

it is way faster in -aps also ... not only low br

also, using the standard preset, 3.96 produces smaller files on most tracks.

Why? Isn't that what is done with Ogg Vorbis? Two, and even three, encoders recommended, varying with the aimed bitrate?

The latest listening test proved the strength of 3.96 on 128kbps. But doubts remain about high bitrates. Recommend both, then.

Confusing or not, I don't think most laymen have HA.org as a source of learning anyway.

It's far from the ideal situation. Many attempts to merge different encoders were done in the past and still recently in order to have only one federative encoder.
For someone used to read HA, collecting many encoders with time... this situation is not complex at all and easy to handle. But for someone new in the audio encoding world, it's probably disturbing and unecessarily complicated.

But I must give you somewhere reason: it's probably not a bad thing to recommand different encoders, if tests are showing that different encoders are better in different bitrate areas.

I agree - the data seems to support two recommended LAME encoder versions right now. Not really a different situation than if (for example only) a Xing MP3 encoder were optimal at low bitrates and LAME at >128 kb/s. One would hope for a unified / universal MP3 encoder recommendation in the future, but for now the data suggests otherwise. MP3 is an important enough format to justify different recommendations for different applications.

It is something to consider. I remember some time ago some people used to concede that Fraunhofer could be a better choice at lower bitrates. Perhaps not the case any more.

If it is a little better and worse at the same time (that is, neither 3.90.3 nor 3.96 can achieve consistent levels of transparency --and I am only interested in APS--), but 3.96 is faster and produces smaller files, consistently, then why is it not the recommended version already?

It achieves pretty much the same as 3.90.3 and it has advantages in speed and size.

I don't know about you, but after reading that thread I have made my choice.

same here ... i really never bother to change from 3.90.3 till 3.96 came out ... but now i'm a 3.96 only guy

Perhaps command-line dependent recommendations could be made.

e.g. 3.96 for aps, 3.90.3 for cbr (just an example) etc...

Looking at (--alt)-preset standard...

To be fair (and this may be one reason for "stagnation"), in half those cases, people took known problem samples for 3.90.3 and threw them at 3.96. In the other half of those cases, people have found new problem samples for 3.96 and thrown them at 3.90.3.

Two years of collecting problem samples for 3.90.3 produced ~four where 3.96 was better. Remember that some of these samples weren't transparent with either encoder, and that there are other problem samples where neither encoder is transparent AND both versions sound similar so people haven't even bothered to post them.

A few weeks of collecting problem samples for 3.96 produced ~four where 3.90.3 was better.

So while there appears to be a roughly 50:50 split in the results, I'd expect more 3.96 problem samples to be found with time.


Having said all that, 3.96 is being developed, while 3.90.3 isn't. If Gabriel and others can solve (some of) the problem samples for 3.96, then it will be going in the right direction - if this happens, we should support it, recommend it, and help it to develop further by making it the recommended version.

If 3.97 takes a different direction, then 3.96 will be a dead end - just like 3.90.3. It's faster, probably about the same quality, but less tested. We should recommend them both - which probably doesn't help anyone!


Looking at other bitrates...

For other bitrates, if 3.96 beats 3.90.3 then it should be recommended - no question about it. If people are confused by "at bitrate X, use encoder Y with settings Z; at bitrate A use encoder B with settings C" then they're probably reading the wrong website.

Cheers,
David.

bump

http://www.rarewares.org/mp3.html

3.96.1 is not yet released

@Gabriel: That's what I thought. Thank you for the confirmation.

Regards,
Madrigal

How is LAME 3.96.1 at Rarewares if it has not been released? Is the compile at Rarewares simply 3.96?

Because it has now been released

look here for latest releases of LAME sources.

Quite the case as to why HA shouldn't have a recomended encoder version rather than just settings for the latest encoder version.

Thank you Gabriel.

I just wanted to verify: Do you mean "--preset standard" or "--alt-preset standard" when you say 3.90.3 and 3.96 are comparable in quality for that setting?

I have also browsed these forums quite a bit, and found conflicting and confusing information. The information for users to make an informed decision between 3.90.3 and 3.96, and which settings to use with each, has not been adequately consolidated.

For 3.90.3 it seems to be established that "--alt-preset standard" will often provide transparent encoding at a decent speed, no? So then 3.96 will give comparable quality at a faster speed using "--alt-preset standard"? Or has the 3.90.3 APS been mapped to "--preset standard" in 3.96? I have read that all the 3.90.3 "alt-presets" have been remapped to the "presets" in 3.96. Is this true? Which VBR algorithm (vbr-old/vbr-new) is used for "--alt-preset standard" and "preset standard" in both 3.90.3 and 3.96? I guess I'm confused as to what settings need to be changed from a user perspective in order to get comparable quality with 3.96 as one would get with 3.90.3 "--alt-preset standard".

Basically, I'd like to make the switch to 3.96 for the possibility of faster encoding and comparable or better quality (while accepting that some samples will end up sounding worse), but I'd like to know what settings to use to get the "equivalent" of 3.90.3 APS, and I'd like to know what kind of "basic" encoding options that's selecting (without detailing the code-level differences from command-line options, etc, just what "kinds" of algorithms/options are being used). I'm sure this information has all been detailed before, but I'm having trouble piecing together all the information regarding 3.90.3 vs. 3.96. and their presets.

Edit: Answering my own questions somewhat, the thread "Newbie Questions" did answer some of what I'm asking above, but I'd still like to hear it corroborated and clarified, and the other portions answered. Any help would be much appreciated.

For 3.96, --alt-preset standard, --preset standard and -V 2 all give identical results.

Alright. 3.90.3 is the best for alt preset standard...

BUT

what for alt preset insane ???

Its the one I use.

Am I still using the correct version.

I'm really not concerned about hdd space.

Just boosted my storage to
620 GB

Thanks!

Lossless is best if you're not concerned about HD space. If you're absolutely insistent on using MP3 at 320 kbps CBR, the best encoder is whichever one sounds the best to you. If you can figure out which one sounds better, please let us know. And give us ABX results.