Newbie question : LAME 3.xx.xx ?
Reply #53 – 2005-06-02 08:37:22
Meh. Everything before 3.96.1 was clearly inferior, if I read the ABX results properly. Clearly inferior? Which samples? Which presets? Give links, please. Many people used 3.91, 3.92 & 3.93, and there were few complaints. 3.95 was selected for Roberto's collective listening test, not because it was "clearly inferior", but simply because quality proved to be superior to 3.90.3.I wasn't referring to recent changes , I was referring to the ones shortly after Dibrom branched development for 3.90.x. I'm talking LAME v3.94, etc. here. I quote your own words:use 3.90.3. Using anything else will give suboptimal results in terms of quality. Every other choice is a decision to trade off quality for speed. Did you ever mention the fact that 3.95, 3.96 & 3.97 are excluded from your first statement? No. You explicitely said that 3.90.3 offers the best quality, without any proof.Please refer to my comment about 3.94. Quality regressed. There wasn't a continual iteration towards perfection, there was a clear regression in the releases after 3.90. Who cares about 3.94 beta? People are interested by 3.96.1 or even 3.97, and are not requesting opinions about 3.90.3 vs 3.94 since two years.ABX is fundamentally based around concepts introduced by predicate and prepositional logic and scientific rigour. Then begin to respect scientific rigour, which supposes to make experimentation after exposing any theory. If you think (dream?) that all improvements made since 3.90.3 were done in order to improve speed with negative impact on quality, that's OK. But then perform listening tests to confirm or infirm this theory. You're not rigourous at all. You explicitely said that you can't detect any difference, but you continue to claim that 3.90.3 have a better quality. It's more than pathetic. There are a lot of people on this board who can't ABX any difference. But they're consequent (call it scientific if you want) enough to not make claims which contradicts their own experience.Like I've stated before, I'll accept that the new versions are superior when they make it onto the Recommended Version page. If they're not there, there's a reason. Yes, and the reason have nothing to do with quality, but is linked to the lack of testers.Until then, let's keep quality debates out of threads where new members are asking for help. Spreading wrong informations doesn't help anybody. You said four things in this topic which are simply invalid or false: 1/ that 3.90.3 offers the best quality 2/ that administrators said that 3.90.3 is better and therefore recommended 3/ that subsequent versions of lame were optimized for speed only 4/ that speed optimisation conduced to lower quality. The 1st one is a simple violation of TOS#8. The 3rd and 4th one are a denigration of Gabriel and Roberto's work, which are constantly requesting from HA.org readers serious listening tests in oder to check the impact on quality of their changes. Therefore, don't say that you're helping the community by answering to people which are looking for advice about LAME quality. 3.90.3 is the recommended version, and therefore should be the one recommended. Yeah, people have opinions, but it is much less confusing if we can just agree that the Recommended Version is the recommended version. When people are looking for the most secured LAME version, then 3.90.3 as recommended version for security should be recommended. That's right. But when people are looking for LAME release offring the best quality (what XP_98 asked for), then 3.90.3 is not the recommended version. Recent tests revealed that 3.97a10 offers better quality than 3.90.3. If you're not agree with it, then prove it, and don't invoke official recommendation which doesn't claim anything about relative 3.90.3 quality, except that "3.96.1 has been proven to be of superior quality for some bitrates where there are no VBR presets for 3.90.3, only ABR " .