what's the impact on encoding quality if I specify '-Z' switch? I use latest lame.

Latest stable lame? or latest lame?

oh sorry, I use LAME 3.94a11 and 3.94a1 sometimes. I always use -q0 switch and no alt-preset.

You'll get worse quality than with alt-presets... but it's your choice after all

I would reccomend use of " -Z 1" with 3.94 you have the ability to directly select the type of noise shaping instead of a toggle on off.

Also you can get better quality out of the non vbr alt presets with switches. Hey its true you can do better on cbr and abr than the alt presets present in the alpha versions.

I wouldn't use 3.94 a1 I would use the latest alpha11 just came out today.

Er, I think it should be pointed out here that alphas shouldn't be used for anything other than testing and/or out of curiosity at what is going on with the development of LAME... alphas can have very serious problems with them a lot of the time, due to their nature (i.e. the initial steps to a modification of an existing stable version).

Using the very first alpha version of an encoder to do anything other than test is... well, extremely inadvisable, to say the least.

The current recommended compiles are 3.90.2 and 3.92. The former is preferred by most people here because it has been more extensively tested due to its age, and is the version officially recommended by this site. See the sticky threads at the top of the MP3-General forum for more info.

Use 3.90.2 and aps for best quality. Also, I believe the -q0 switch isn't recommend because its worse quality then -q2.

It's also worth noting that 3.90.2 is slightly faster than 3.92. Although not much.

Lame v3.90.2, --alt-preset standard, Athlon XP 1700+ (@1511 Mhz)

Lame v3.92, --alt-preset standard, Athlon XP 1700+ (@1511 Mhz)


Tested track was "Everything But The Girl - Missing".

3.90.2 generates bigger files, but I belive this issue has been debated extensively elsewhere on the forum.

I have been reading many many posts and I must say that this is pretty confusing. I guess, everyone says that 3.90.2 is the best. So...

1.) What happened with LAME releases after 3.90.2? I supposed that developers enhanced something, rather than made it worse...?

2.) Why is -q0 worse than -q2? Please, just do not tell me that it's just because of the bigger files...

3.) My command line looks like this: 'LAME -q0 -Z -b256 infile.wav outfile.mp3'. I encode CBR because I noticed somewhat "cleaner" sound generaly.

4.) My audio equipment is SB Live! 5.1, Technics SA-AX720 (5x100W) receiver (it was top model 3 years ago) and 2x120W 3-way Pioneer speakers.

5.) ANY comment is well appriciated.

Thank you.

1, 3.90.2 is released by Dibrom. He added alt presets swithes in this version which had been tested many times and proved safe.
2, q0 is used to be a testing switch. q2 is faster and you can't lose any quality. Maybe latest lame verson changed this.
3, You just hear what you want to hear. You'd better do some ABX test.

You can find these answers from the sticky threads.

http://www.hydrogenaudio.org/forums/index....=ST&f=15&t=3836

http://www.hydrogenaudio.org/forums/index....=ST&f=16&t=4336

http://www.hydrogenaudio.org/forums/index....=ST&f=15&t=4544

-q0: It should be working reasonably ok in alpha11. Don't use it in non-alphas, since Takehiro has said (and I've tested) that it's somewhat buggy in non-alphas. However, in alphas it uses substep noise-shaping which is mainly meant for low-bitrates. If you don't specify -q value in alphas, -q3 will be used, which is the same as the tested -q2/h in non-alphas.

-Z: If you are using GPsyhco like in this case, it by default uses noise-shaping 1. By using -Z you switch it to noise-shaping 2, which is not working well with GPsycho. Not recommended here with that line. Don't just add switches you see somewhere and think it will increase quality. Depending whether nspsytune or gpsycho is used, the effect of -Z can be completely the opposite...

Well, this works only with nspsytune. Try using -Z1 or -Z2 with gpsycho. You get exactly the opposite you define..

We changed the -q mapping in alpha.
Substep is enabled from q2, and at q0 your are using a full huffman search (awfully slow).

Eeh, so it was changed few weeks ago to this?

-q3: same.
-q2: -q3 + substep shaping method 2 and increase max_age.
-q1: -q2 + only 1 sfb to amplify in the inner loop.
-q0: -q1 + best huffman in the inner loop.

So you are now using --substep 2 in -q0,-q1 and -q2?

Anyway, as people can see, things are constantly changing with alpha. The default -q3 (which is the old -q2/h in non-alphas) should be safe in alphas, but of course anything can happen.

Perhaps the warning displayed by alpha versions is not big enough?

What is gpsycho?
So will aps or ape with -Z (LAME 3.90.2) be of better quality or worse than simple aps or ape?

Can -Z hurt quality with aps and ape (LAME 3.90.2)?

Gpsycho is the default psycoacoustic model used by Lame. Nspsytune which adds Naoki Shibata's tweakings to the psychomodel have to be enabled with --nspsytune swithch.

I already told you few days ago the answer to your second question.. No, -Z does not hurt 3.90.2-3.93.1 APS or 3.90.2-3.93.1 ape, on the contrary...but do not use it with 3.90.2-3.93.1 insane profile, since it already uses noise shaping 1................
Use the --verbose. Noise shaping 1 is always the safer choice whether using gpsycho or nspsytune especially with vbr.

Sorry, I didn't mean to ask the same question again, but I got a bit confused
reading this thread.



U mean like this?

--alt-preset insane --verbose

Heh that is a little messed up john.

Perhaps it would be prudent to make the same change for gpsycho that was made for nspsytune. Perhaps plans are already in place to update nspsytune and remove gpspycho? I heard dibrom and gabriel mention this perhaps takehiro can shed some light on the subject.