graph shows 320 kbps LAME 'never achieves transparency'? Options

[krabapple]

Someone is waving this in my face at another forum,after I asserted that people who can tell modern 320kbs LAME encodes from source in ABX constitute the a tiny minority of listeners, and even they may require 'killer samples' , rather than being able to do it all the time, e.g., with a random mix of music.

You talk about LAME. Let's look at that: http://wiki.hydrogenaudio.org/index.php?title=Lame

Here is a graph from that page:


It shows that it *never* achieves transparency regardless of data rate.

I called bullsh*t on that rhetoric (e.g., on the very same wiki page, we see "-V0 (~245 kbps), -V1 (~225 kbps), -V2 (~190 kbps) or -V3 (~175 kbps) are recommended. These settings will normally produce transparent encoding (transparent = most people can't distinguish the MP3 from the original in an ABX blind test). Audible differences between these presets exist, but are rare. "

IMO, this graph is not evidence that 320 'never achieves transparency'. I know some people can ABX 320kbps. That is not the issue -- I have never claimed 320 is transparent to EVERYONE, all the time. Just that it's a rare ability, typically requiring training to hear specific artifacts, and possibly requiring carefully-chosen samples to reveal them.

The graph also lacks error/variation bars -- where's from, btw?

Dude (who is a former colleague of JJ) refuses to come over here and argue his case. I told him I'd bring it here for him. What say you, LAME developers and users?


full post:
http://www.avsforum.com/avs-vb/showpost.ph...p;postcount=348

This post has been edited by krabapple: Aug 26 2011, 19:33

[db1989]

Yep, the p-value is the probability that the observed result could otherwise have been obtained by chance. A result over the chosen significance level (usually 0.05) leads one to (not accept, but) fail to reject the null hypothesis, which usually proposes no effect of whichever treatment is being investigated.

I too am not sure how one can demonstrate that there is not an audible difference. One cannot confirm the null hypothesis (i.e. prove absence of an effect), only fail to reject it. Thus, transparency could be argued if multiple users failed to achieve statistically significant p-values, but not the converse.