LAME 3.90.3 is still the recommended encoder at Hydrogen Audio.

However there seems to be afiar amount of work done now on LAME 3.95.1 http://www.hydrogenaudio.org/forums/index....showtopic=18091. The new V presets appear promising.

My question: is 3.95.1 far enough along where the encoder and its V presets are being used by peole to encode their music collections, or is it still in testing mode only?

And what are the key advantages people are finding with it that would cause them to use 3.95.1 rather than 3.90.3. I saw a thread where file size was suggested but that seems to still be open for debate. Is it encoding speed only?

And how does 3.96 figure in to the equation?

Again, I do not want to confuse anyone - 3.90.3 is still the recommended encoder on the LAME settings thread

Ehm... your question confuses me.
You either go the HA.org road and use 3.90.3 or follow the path of the official LAME developers and use 3.96.1.
There's no reason to use 3.95.1, since 3.96.1 is just its logical successor.

Maybe this post can help you.

There are huge improvements from 3.90 to 3.96, I think everyone should be using it now. 3.90.3 is counterproductive & obsolete.

Just read the lame change log & the lame regression thread.

This is obviously your personal opinion, it does not respect the whole community.
Quality speaking, 3.90.3 is far from being "counterproductive & obsolete".
The topic is about 3.95.1. There is no reason to use it, you have to choose between 3.96.1 (only for -V5 --athaa-sensitivity 1 in some cases) and 3.90.3.

If you are encoding at high bitrate-ranges, then:

- if safety and guaranteed stability (quality-wise) is most important to you, then use 3.90.3
- if speed and a slightly better quality/filesize-ratio is most important to you, then use 3.96.1

If you are encoding at mid-low bitrate-ranges, then:
- use 3.96.1

i guess thats as simple as one can put it,
- Lyx

Hi guys,

Do any of the 3.90.3 -> 3.96.1 changes affect api? I couldn't find any info on this and abx'ing is difficult and time consuming at this setting.

Of course api is affected by the changes in LAME, but nobody has bothered testing it yet for the reasons you mentioned (and the general lack of interest in 320kbps cbr).
Odds are that you won't hear any difference at all.

Yeah I know I've only recently started abx'ing and I'm still getting used to my new equipment (Total BitHead & ER4-S). On some music I can't even abx 128cbr.
I have been reading the threads on this forum for two-ish years, so I'm aware of the situation, but 4.4x compression is enough for me (given the amount of music I own and the size of portable I'm looking at getting). Of course if I stick with aps I have more room for expansion and/or could get a cheaper/smaller portable, but I'm not ready to make that decision yet (still contemplating lossless also).

Hope I sound just a little bit less like a newbie now

If you're considering other formats, you might want to take a look at Vorbis and especially MPC too.

The reason I say it's counterproductive is because if everyone uses the old lame code it prevents finding problem samples & ironing out the bugs of the latest code which is in development.

I say it's obsolete because lame 3.96 achieves equal quality at a lower bit-rate.

But of course the Hydrogenaudio Recommendation is in the recommended lame compile thread which has been dead for like over a year now.

Thanks.. I'm leaning towards the iRiver iHP-120/140 at the mo mostly because of the mass storage support (so I can just use it as a harddrive and use Foobar at work), and an FM tuner would be nice.

I haven't started to abx Vorbis yet - I'm slightly put off by comments about high bitrates being unsupported on portables. I need to investigate further, especially since some comments seem to suggest it now sounds better than mpc.

Hmm... some quick rough calculations tells me that for a 20G drive:
wav = 30 albums (battery killer!)
ll = 60 albums (ll=some lossless format, unsupported!)
api = 140 albums
aps = 230 albums
aps+mpc = 110 albums? (worst case scenario of -q 7)
aps+ll = 50 albums

100 albums is enough for me (on a portable), so quite a few possibilities.
Far too much abx'ing lies ahead

Erm... sorry for taking the thread OT.

This is unfortunately true but cannot be the reason to use 3.96.1. The final user wants quality guaranties.

The 3.96 abr mode is much inferior to the old one (see previous tests).
Someone could prefer the new vbr presets at ~128 kbps but people with good HF hearing may find ringing/ATH problems annoying. With high bitrate presets the old --aps is still more tested to give near transparent results.

You're forgetting the recent tests done before the 3.96 stable release.

edit: That said, we are OT in this thread. This discussion can last for a long time uselessly.

Are the preset names for 3.96.1 the same as for 3.90.3 or are they different?

Never mind, found the 3.90.3 vs. 3.96 thread.

Which really didn't prove anything...

I give you a advise that you had best use lame 3.96.1 which convert speed is very fast .

I totally agree. It is very counterproductive to have a recommended version not the latest version. Everyone will always fall back onto the arguement that the older version is better because of the mass testing that was applied to it. If we all keep recommending 3.90.3 and not newer versions I'm afraid we are slowing development and actually hindering the potential of the newer versions. Nobody seems to want to test the newer versions because 3.90.3 has already been tested and proven, so lets all fall back on that and not care about advancing the format. We may as well tell Gabriel and the other LAME devs that we don't care about newer versions. Do we all not want LAME mp3 to move forward?

I purpose we ALWAYS recommend the latest version of LAME and ALSO have a recommended “older” “stable” version. Recommending the latest version will spur further advancement of the format rather than subjugating it to a relic.

I think the "it is counterproductive to development" is irrelevant. The HA recommended settings/encoders are meant to give the >user< the best (in overall - there are multiple factors) encoder/settings. As far as i know it is NOT meant to advance development of encoders. Because of the nature of HA.org, i would also say that the "HA-recommended" threads are also meant to educate and evangelize the masses - or in other words: to spread the gained wisdom.

This imho is a much bigger argument against having 3.90.3 as the recommended version. Staying at 3.90.3 after 3.96 got out (and received significant(not large-scale) testing) was tactically a big mistake imho. The reason for this is the following:

When LAME 3.93 was the current lame version, it was more easy to advice users to use 3.90.3 instead - because the version difference wasn't that big, AND 3.93 had significant regressions...... at LAME 3.94 it was still possible...... at LAME 3.95 it became difficult, because years of development had passed and 3.90.3 was "old" (that does not mean worse!) - also, initial tests indicated that 3.95 was catching up with 3.90.3 and seemed to not be much worse - on medium bitrate-ranges it even appeared to be better. The enourmous speed-improvement also should be taken into account.... then came 3.96 which again received significant tests which indicated that its not worse than 3.90.3 quality-wise - and again - at medium bitrates even appeared to be better than 3.90.3.

Why do the above things matter for tactical reasons? Well, it is moronic to educate the masses and spread recommendations, if your voice has little significance. To make users use an older version, there need to be significant reasons for doing so. This has not only something to do with quality: the bigger the versionnumber gap, the more difficult it becomes to convice the average user to use an older version instead. The faster the new version becomes, the more difficult to convice the average user to use an older version instead / the lower the risks of the current version, the more difficult to make them use an old version / the smaller the quality gap the less they will care, and so on.....

So, with the above things in mind, it is clear imho that the significance and "respect" for ha-recommendations will become lower, if a version like lame 3.96.1 is out and you still stay on the "safety-bandwaggon" of 3.90.3 - because the "gap" between the current version and 3.90.3 has just become too narrow - and in quite some aspects, 3.96.1 even overtook 3.90.3. To newcomers and average users this makes ha.org recommendations appear like some obsolete advices of uber-conservatives who aren't up-to-date with current developments. And exactly this has already begun to happen...... 3.96.1 and 3.90.3 have split ha-org into two user-groups, when it comes to mp3. And when someone asks which version he should use, he doesn't get a clear answer anymore, but instead "oh, well, "theoretically" you should use this, but in some cases you may want to use this".

So, yes - from a quality/safety-purists point of view, 3.90.3 is still the best version - but as soon as you take more factors into account and "see the whole picture", 3.96.1 clearly is the better choice for a ha.org-recommendation. After all, what are ha-recommendations worth, if they have a low significance/weight/impact on the users? right, not much.

- Lyx

edit: what it comes down to is: keeping 3.90.3 as the recommended version now, is counterproductive to the significance/weight of the voice of ha.org - so in other words, ha.org is loosing power to educate newcomers and the masses.

There are two problems (at least):

- first, evidence of 3.90.3 superiority are very limited. Quality is better on some samples, worse on other... I could find plenty of samples where vorbis RC2 sounds better than 1.01, but is it enough to recommand the very old vorbis encoder?

- the "recommanded encoder" thread doesn't give the reasons of 3.90.3 choice IIRC. I read some reply in french forums recommanding 3.90.3 over newer release but for fanciful reasons (corruption of files encoded with 3.96, version optimized for speed only, etc...). In other word, some people are trying to guess why lame >3.90 must be avoided, and are therefore spreading wrong informations. It's exactly the opposite of HA purpose. What a paradox!

Yes, this is true. It's prompted me to answer anyone who asks what LAME version to use to just say test both. I believe it is much more educational to have a new user to perform ABX tests on any codec than just a thread of barfing out encoders and versions.

My take on things:

The most recent versions should not be the recommended versions. However, it's perfectly fine to add a note under the 3.90.3 recommendation stating that any help in testing the newer versions would be appreciated.

However, the quickest and easiest way for 3.90.3 to be replaced with a recent version is for the LAME developers to show an organized effort in properly testing for quality. If they had some sort of system in place for conducting regression tests and some systematic way of being able to certify builds as being "stable", we wouldn't have this problem. But this kind of thing has been discussed to death already, and I doubt it's ever going to happen.

Bottom line though is that this isn't really an HA problem, it's a LAME problem.

The recommended versions thread can certainly be updated with some qualifications as to why certain versions are recommended and others aren't, but I don't think we need to really make any changes until there are clear reasons to do so. Unfortunately, even after multiple revisions since 3.90.3, as far as quality at high bitrate vbr is concerned there still doesn't seem to be much of one...

That's what the LAME Versions and alphas thread is for.

Additionally I added a note to the recommended commandlines thread refering to the explanation of the new -V preset system.

for clarification and avoiding misunderstandings:

I'm not advocating to -always- recommend the latest version. What i was trying to explain was that imho 3.96.1 has just become too good and close to 3.90.3 to recommend 3.90.3 anymore (again, not only taking quality into account but all factors). Small details aside, the only argument we have left to advocate 3.90.3 is that it received larger testing - in all other regards 3.96 appears to either be equal or better. Since its very possible that there will never again be such a large-scale test of mp3, even the "more well tested"-argument looses significance, since it implies that 3.90.3 maybe can never be replaced no matter how good newer versions become.

- Lyx

FYI...