you sure you used track --scale, or did you apply the album --scale?

Yeh, it looks like the Rg values LAME is spitting out here are not correct either - my afformentioned Cure album has some tracks at -6dB track after applying album, but the wavegain analysis had all the tracks within a dB of eachother before encoding.

<edit>
I just realized, LAME 3.95 is using the old value of 83dB, instead of the 89dB everyone and their dog uses today. This should totally be fixed!

Excellent.

Some light coming soon!

I don´t test much only some samples. aps 3.95 adds an artifact "plop"
at about second 3 in sophia2. 3.90.3 doesn´t.

Wombat

I need a little advice on how I can help with testing. My problem is to chose the material to test. Should I select tracks by random out of my 400 CD's, or should I download testsamples from a testsample-archive. The first choice seems to me totally useless because I don't know on which part of music I have to pay my attention too. I don't have so much experience in ABXing music. I'm sure, that Icannot tell the difference between 9.90.3 and 9.95 in 99% of the cases, so is there a strategy to chose the right samples?

There is going to be a fix 3.95.1 release later today.

dev0

just a quick question; why does replaygain reduce the bitrate? It stays the ame quality, right? Is it something like a 2-pass encoding? How do I turn that on? I always thought that replaygain changed the volume of a mp3 file

You let wavegain (availible on rarewarez) calculate the album gain (looks something like -7.34 dB, factor 0.6436).
Write down or remember the 0.xxxx factor, then encode all wav for that album with "--preset standard --scale 0.xxxx"

Why it reduces bitrate.. search the forum for "sfb 21"
Same quality, well.. I certainly hope so In theory yes

The end result should be the same as if you mp3gained the album after encode, just with slightly lower bitrate.

Oh yeah you can also let wavegain manipulate the wav files directly and you can forget about the --scale stuff, but I dont like that because my harddrive sucks and writing 600 MB takes long

Anyone else noticed this? I haven't downloaded 3.95 yet, so I can't tell.

How are useful listning tests set up? Is it better if this version is going to be compared to 3.90.3 with the same encoding settings, or to compare different settings with wavs, looking for things that don't encode that well, disregarding how 3.90.3 would have done? 3.95 seems to give lower bitrates so I guess it makes sence comparing it to 3.90.3 with the same settings, since 3.90.3 is much like a reference now?

Is it of any use if I try to describe the differences I hear when "succesfully" ABX-ing samples, since others may not hear it that way?

Just want to precise that lame 3.94 (didn't test 3.95 yet) is not bitrate friendly with classical music compared to 3.90.3 : 15...20 kbps more on average (difference is sometimes +50 kbps with some tracks).

All those bitrate/speed comparisons are useless as long as you don't provide any information about the quality (especially compared to the currently recommended 3.90.3).
dev0

Yes, testing is needed.

As I statet here, there should be an advice what testing makes sense, to get the community involved a little better.

well I don't think well presented speed and average bitrate comparisons are "useless".

certainly, everyone wants to know if the .95 aps is safe to use... but that is ultra-difficult to prove. I personally have good faith in the lame devteam, and am willing to benefit from the speed increase and bitrate decrease until .95 is proven to be inferior on aps

thank you very much for your answer, sony666! However, I wonder why this hasn't been implemented in LAME yet. I mean if it saves 23kbps isn't that great?!

In using --scale, I found that my loudest album (St. Anger) required --scale .85 to encode without clipping. I just use that setting for everything.

That seems to keep everything under 0db. I let Foobar2000 calculate the RG values after encoding. This gives me an encoded file that will not clip on hardware, and consistent RG volume when using Foobar2000.

This might be the wrong approach, but it works well for me.

Edit: Sorry if this strayed off-topic. Please split this thread off it it continues down this path.