hi,

is there any known sample where Lame 3.90.3 beats Lame 3.93.1?

Some samples like erhu maybe. But comparison isn't fair : you had to compare 3.90.3 --alt-preset standard and 3.93.1 --preset standard -Z.

HA recomends 3.90.3, but it look like nobody verified that 3.90.3 --alt-preset standard sounds superior to 3.93.1 --preset standard -Z.

You're missing the point which is that a huge amount of testing and tuning went into the creation of 3.90.2 (3.90.3 is just 3.90.2 with -Z automatically used for the standard presets) and this is the yardstick used for comparison. There has not been the same exhaustive testing applied to any other version, 3.93.1 included, to know the answer.

By encuraging the usage of an old version of a software, you are effectively defeating the developement of the software. Using the argument that the old version is more tested than the latest, will only make that situation stay. This community should be encuraged always to use the latest version (but comparing with older versions), and send feedback to the developers.

I believe there are other reasons to use the the latest version as well:
- 3.93.1 is 7-8 month old now, and is actually very well tested.
- Generally faster than 3.90.3
- Better optimized for for different bitrates (lower): --preset <bitrate>

There should be a major ABC/HR test concluded between these two on all samples,
but I think most of the people around here wouldn't be interested.

If anybody has any ABX results, please post them in the thread...

The recommendation is just what it is - you might not heed it and your results will still be good,
but probably not as good as with the recommended version.
But if you'll find any problem, be sure to test with 3.90.3 too.

Just for the sake of the higher version number? You're a victim of Versionitis, my friend.



What we're talking about when we say "testing" is, specifically, testing the performance of the alt-presets. And these certainly have not been as well tested for 3.93.1 as for the versions recommended by HA, because Dibrom no longer maintains them for the current LAME versions (as a result of a major clash with some of the resposible LAME developers, but that's a different story...).

Some interesting threads:

The discussion (mostly by Dibrom and Gabriel) about the alt-presets in 3.93:
http://www.hydrogenaudio.org/forums/index.php?showtopic=3836

Continued here:
http://www.hydrogenaudio.org/forums/index.php?showtopic=3920

The release of 3.93.1:
http://www.hydrogenaudio.org/forums/index.php?showtopic=4614

Reading these threads can learn you a lot about how things have evolved with LAME.

Er... this was never actually the problem. The problem was in getting the developers to "do the right thing" and actually address the points raised during feedback.

To this day, there are still many known quality issues with LAME which have never seriously been looked at, or at least publically acknowledged, by most of the LAME developers.

To be brutally honest, I think LAME (or at least the development of the project as we knew it) is pretty much dead. I know that Takehiro is working on the supposed LAME 4, but with the way the development structure is headed (or has been heading for some time), I have a hard time imagining this to end up being the successful release that so many people are probably wishing or expecting that it would be. Anymore, there seems to be little motivation for the existing developers to seriously push things forward, and without a guiding force (last I heard, Mark, who was already rather uninvolved long enough for some time before stepping down, was no longer the maintainer, and Alexander was only the "pseudo" maintainer or something), I don't really think things will ever seriously get back on track. I also think that there are some serious issues with inertia and fear of change held by some of the still influential developers associated with the project (which was a large contributing factor in my discontinuing of my involvement with the project), which certainly do not make things easy in this regard. As if this weren't enough, Takehiro, who seems to be the only one really interested in seriously reworking things, apparently isn't too keen on people testing his code and providing feedback, at least not in the HA fashion. IIRC, There was actually a thread here awhile ago where he got quite annoyed at john33 for posting some of his early work on the new version.

I could be wrong about all of this of course, and I hope that I actually am, but honestly, I'd be rather surprised if I were.

Thanks for those answers and links. Makes me understand the complexity of managing this project, and that an "ideal" developement path not always can be followed (although, beeing a developer myself, I think my above points are valid for most projects).

Even though there are known quality issues with LAME, isn't it still fair to say that the further potential for this software, given the limitation of mp3, is really small compared to what is achieved so far? For my needs, LAME is perfect already.

thanks for the replys and I hope the split of Lame in HA and official brach will be stopped at the next release 3.94.

B)

It's funny how things change.

LAME, MPC and even Vorbis seem terminal. The current versions are hardly improved over compiles from 1 year ago. These encoders have been "floating" for some time. I am not in a position to assign blame but I will say it is shocking how much the spout has been twisted from a gusher to a trickle.