which is better? lame3.96b2 or lame3.90.3?

List of recommended LAME compiles
"This list will be kept up to date with the current LAME compiles that are recommended for use for optimal quality and will keep track of any developmental compiles released on this site.

The currently recommended LAME version:

--Download LAME 3.90.3--"

3.90.3 is finished code that is thoroughly tested and refined while the beta is obviously an ongoing project subject to change and mostly untested

lame 3.96b2 is a beta version, and should not be used for other purposes than testing. There is however a stable 3.96 release. But 3.90.3 is still the recommended compile so...

may be , 3.96b2 is recommended in another web site!

http://www.hydrogenaudio.org/forums/index....howtopic=19881&

It is still beta, and a site recommending a beta build when there is a new stable release, cannot be trusted. There are probably lots of sites out there claiming to know what is the best version, but none of them can come close to the knowledge of this board. Just the fact that most (if not all) developers of open standard codecs are member of this board proves why.

Some samples are better with 3.96, some with 3.90.3. No clear winner.

I swear on LAME 3.96 final

HydrogenAudio recommends MPC 1.14 beta while 1.1 is stable. It does not make it untrustable.



I don't think this kind of arguments will convince anyone. Lqn168 saw that 3.90.3 was "the best", then that 3.96b2 was "the best". So he is confused and asks, now we tell him that "we" are "the best". What will tell the other site ?



That's a better argument, though not proving much in itself


This does not help either.

Follow the link given by Dev0. You can reach the test results between 3.90.3 and 3.96 from there.

I would also add:
http://www.hydrogenaudio.org/forums/index....howtopic=20715&

Regarding quality, it seems that there is no clear winner between 3.90.3 and 3.96 when using preset standard.
Regarding VBR when using lower bitrates (like -V5), 3.96 has a clear advantage.
Regarding speed, 3.96 has a clear advantage.

If 3.96 and 3.90.3 are tied at high-bitrates (--preset standard).... but 3.96 is better than 3.90.3 most of the time at medium-low bitrates..... and 3.96 is faster than 3.90.3...... why is 3.90.3 still the recommended version then?

I think Lyx has a good point..Why is 3.96 not the Offical Recommended Hydrogenaudio version?
Cheers,
Scott

That comparison is not fair. There is a difference in recommending 3.96 beta 2 when 3.96 stable is out, and recommending 1.14 beta (which has been used for very long time, and can be considered pretty stable ?) while waiting for the next stable release.

As for the rest of your comments, yeah they (mine) weren't to smart, I have to admit that. But I still think that the fact that most developers are members proves something. Still, not one of my greatest hits on this board...

The answer which will probably come up is "oh, but 3.96 wasn't tested as much as 3.90.3".

LAME will probably never again be tested on such a large-scale as 3.90.3 (unless maybe when 4.00 comes out). Does that mean that the recommended version has to stay 3.90.3 forever?

Isn't hydrogenaudio.org in that case risking to provide "outdated" information? Especially concerning the recommended-version thread and the mp3-faq. This would not just be of disadvantage for the community itself: if ha.org becomes known for preaching outdated truths, then its impact outside of this community will fade.

Being cautious and skeptic is a good thing - but it's the dose which makes the poison.

- Lyx

I think the reason is that it is considered to need more testing. If I recall correctly there were comments that all testing so far was done on samples where 3.90.3 had problems. Obviously 3.96 has improved on those, but that does not mean that 3.96 may have become worse on other samples.

But I too agree that it would be a good thing if 3.96 could be made the new recommended build. I have seen on various threads in this board that people are starting to recommend 3.96 even though 3.90.3 is till the official build. It does look a bit confusing.

EDIT: Lyx was a bit faster than me...

In my opinion, the first step towards a serious effort to replace 3.90.3 should begin by making sure that 3.96 does not regress in quality on any known problem samples versus 3.90.3. If we have that, and there are very clear benefits in other areas (speed, lower bitrates), then the decision will be easy.

With my experience in experimenting with LAME, I don't think it should be impossible to achieve this goal either really. It might just take a couple bouts of short lived but intensive tuning, working with some of the people on the boards here.

This has already been done in the thread above linked by Gabriel, and it has been shown that 3.96 preset standard has regressed on some problem samples. I think the people that were participating in this effort saw this and just stopped testing realizing it was futile, myself included. So any effort on regression testing must be directed towards a newer version (if we are to compare preset standard).

I'm inclined to use 3.96 V1. It has improved quality a lot for me on every (alt) preset standard problem sample (both 3.96 and 3.90.3) I tested, and the increase in bit rate is not too substantial--no where near the bloat of (alt) preset extreme. On some samples it is lower than 3.90.3 aps. In ABX testing 3.90.3 vs 3.96, I just found too many cases where neither 3.90.3 aps or 3.96 ps is transparent enough for me, most recently in this sample:

http://www.hydrogenaudio.org/forums/index....topic=22241&hl=

So for me, 3.96 V1 seems like the ideal tradeoff between quality and size, and as a bonus its faster than 3.90.3.

I can understand if people are not willing to allow a bit rate increase in the recommended setting. So perhaps further tuning on either preset standard or V1 will be needed. But I also remember people balked at the bit rate increase from r3mix to alt preset standard. At least its something to think about.

Then this would be an easy choice. But if there are regressions on some samples while there are also progressions on some other samples?

Well then someone needs to decide if the progressions are worth the regressions.

In my opinion, if there already exists the possibility for better quality (by the progressions), then every effort should be made to close up the gap in the area where it doesn't perform as well. Like I said, I think if these areas were really focused on seriously (perhaps to the exception of everything else, until this has been achieved), it shouldn't be too hard.

Most of the alt-preset stuff was just a hack. If it's possible for me to have improved quality in that way, then it would seem possible to do something much better with a more focused effort.

The rational behind the recommendation to use 3.90.3 instead of newer versions is that 3.90.3 has been very thoroughly tested (i.e. listening tests, etc.) and newer versions have not been tested to as great of an extent. Frankly, that rational never really sat to well with me because it just seemed to vague. Had the rational for sticking with 3.90.3 been something like "stick with 3.90.3 because newer versions have also been tested and have been proven to be inferior to 3.90.3" then I would be very comfortable with that.

That being said, you really would have to do your own listening tests and judge for yourself. I generally try to refrain from stating my findings in listening tests because I feel that some readers may put too much weight in it and not bother conducting their own listening tests. But since I've noticed that previous posts in this thread seem to indicate that the newer versions perform (quality wise) as good as (or better than) 3.90.3 at lower bitrates but that 3.90.3 performs (quality wise) better at higher bitrates, I will say that I definately agree with that statement, based on my own listening tests.

So, for me at least, it really comes down to what rate I plan to encode at. If "--alt-preset standard" or higher, I would definately stick with 3.90.3 as I've never been able to distinguish MP3 from original CD with "--alt-preset extreme" or higher on any sample and have only been able to distinguish MP3 from original CD with "--alt-preset standard" on a small handful of samples. But with newer versions, I've been able to distinguish MP3 from original CD with "--alt-preset standard" on some of the same samples that I wasn't able to distinguish between when using 3.90.3.

As far as bitrates below "--alt-preset standard", I am much more often able to distinguish between MP3 and original CD with both 3.90.3 and with the newer versions. So, in that case, it really doesn't matter anyway so the fact that newer versions may encode faster and/or produce smaller files may be more of a reason to consider using newer versions when encoding at these lower rates. Also, at these lower rates, even though I was more often able to distinguish MP3 from original CD in both cases (i.e. 3.90.3 and also in newer versions), I felt that, in general, the samples encoded with the newer versions actually sounded better than 3.90.3, but that is pretty subjective and it could have just been my imagination.

But, like I said before, you really need to perform your own listening tests at various bitrates with various samples with each version. My tests were in no way scientific or, for that matter, even very thorough. I'm talking about a pretty small number of samples (mostly 80's music and maybe one or two rap (Beastie Boys, of course) samples and maybe one or two heavy metal (Metallica) samples and even a few classical samples just for good measure) and only found my own best method of conducting my listening tests very recently (see below for what worked for me). I didn't bother much with the "normally problematic" samples like FatBoy.wav, as I feel that stuff like that is really more of an anomaly and does not accurately reflect the results I could expect in my own music that I normally listen to.

On a side note, since you appear to be relatively new to this stuff (based on your post count and your join date), just like I am, I can tell you that I had a heck of a time getting used to performing listening tests and feeling confident in my results. In the beginning, and up until pretty recently, I wasn't even able to distinguish 128kbps encoded files from the originals. A few tricks that I've found to help, for me at least, are:

1. Don't try to test with entire songs. Rather, take a few seconds of the song where you think you may hear a difference and test with part of the song only.
2. Get yourself a good pair of headphones (I'm using Sony MDR-7506).
3. Close your eyes while listening to each sample when conducting your listening test (this was huge, and recent, for me - although it may not work for you).

Good luck. Have fun!