Dear forum,

Is the r3mix lame vbr setting is the best so far ?
best=good sound and small file size ?

Thank you.

The --r3mix setting does provide good quality at a fairly low bitrate, but depending on what bitrate you are shooting for, you may also wish to try --dm-preset standard. This setting provides higher quality, but at a slightly higher bitrate.

In a recent blind test with almost 40 participants it has been found to provide around the highest quality overall across a wide variety of switches, --r3mix included.

There are many critical samples which --dm-preset standard performs better on as well, such as test clips like 2nd_vent and serioustrouble.

Hmm, the results of that test are still under discussion (actually, I'm waiting for ff123 to finish his analysis tool with the nonparametric Tukey HSD test )

With the (confirmed correct) analysis we have so far the Wilcoxon test indicates what you say (except for CBR256, which was the same quality), but it has to be noted the expectation is that at least _some_ of the conclusions from that analysis are actually incorrect (an unforunate byproduct of the fact that there were so many samples in the test). If one looks at the individual results, the r3mix vs Dibrom result is quite 'safe' (< 0.5%. chance for the xtreme preset, <1.1% for the standard preset)

If this all sounds confusing, don't worry, it is. I think we all got a cold shower on that test as far as statistics are concerned

--
GCP

Which is why I didn't say that --dm-preset standard was the best, but was around the best. I of course take the test results with a grain of salt and I'm not talking about any absolute conclusions really, only stating what I think at this point is pretty much established (and which you seem to support with your statement) as the obvious.

All of that aside though, many other results from individuals performing abx tests have confirmed most of this to be true as well.

Dear ookzDVD,

I participated in the last listening test and let me say that the r3mix setting stinks. I correctly identified the worst sample (192), the second worst (r3mix), and as with other people I correctly identified dm standard and dm xtreme as being very high quality.

Now that we don't have to cower under communist rule like on that old board... let me say that you will get much better soundquality at about the samefilesize if you just use dm-preset standard.

But as always best thing to do is just take some music files you know well and encode them in r3mix and dm-preste standard and then do some blind listening tests and see if you can tell the difference. Just don't do this will too many files all at once.

RD

Now this has got me interested. I didn't pay any attention to the test, and after that big flamewar in the results thread I lost track of the actual results, too.
But now you've said that... hmm, can I still get the test samples somewhere? I've never done intensive testing with --r3mix, and when I did some very brief listening tests I couldn't tell the difference to the original - I'd like to see this for myself. (Hell, I'm a bit worried now, as I have encoded masses of music with this setting! )

CU
Dominic

Edit: Umm, I _can_ actually see the ranking of the codecs used from ff123's post...

hmm, if I remember well, the MPC preset used for the test was something like a "-insane dunno-the-rest" which gives us too high avg bitrate, wasn't it?

Was that really necessary!?

I mean what didn't they used "-xtreme..." at 2 or 3 hundreds of kylobits per second which I'm sure could also have beaten MP3?

The purpose of this really-insane MPC sample was to be some sort of a 'control'. Theoretically, it has to be the best sounding sample, therefore if there are any listeners who do not rank this sample as the best sounding sample, we can consider doubting the validity of the results this listener submitted.

I splitted this thread and placed the statistical method speculation messages to the Mayhem General forum under the topic of "Statistical Methods for Listening Tests".

http://www.hydrogenaudio.org/forums/showth...s=&threadid=108

It was getting pretty off topic (R3mix VBR settings), and I felt the statistical method speculation needed a thread of its own... Sorry for any inconvinience. Feel feel to complain if the split wasn't done correctly. Thanks