Downloaded a LAME 3.96 source from mp3dev.org.

While compiling in VC5 (which might be the first problem as the source is apparently intended for VC6) i got approx. 30 warnings. Some of them were just about comparing signed and unsigned values (probably not causing the problem). But some of them stated an integral constant overflow which was more alarming. Others said that some information might be lost while converting "int" to "short" (or something like that).

The complete build was successful, but lame.exe produced lesser quality MP3s (compared to the 3.95 version).

The sample used for testing can be had here:

http://www.volny.cz/martel

When encoded with "lame396.exe -b192 -q0" it sounded up somewhat strange, containing audible artifacts.
Then tried lame395 with the same commandline and it sounded up much better...
(still not that good as MP+)

Could it be the problem of my custom build or is it caused by changes to the source since 3.95?

Just don't tell me it's a problem of the song's genre...

To know if this is a compilation issue, you should also try with a version that you did not compiled yourself.

I am not hearing anything particular on this song.

When i browsed your homepage, i found only the 3.94 win executable.
Can you give me some link to the 3.96 win32 executable?
(also tried www.mp3dev.org, but it seem to be down)

rarewares.org

Thanks. Going to try immediately.

I don't want to disappoint you, but the problems i described before exist even for the downloaded 3.96 exe (from the rarewares).
But i found out more about the problem. It appears with the -q 0 setting only which should use the most precise algorithms but, instead, it ruins the quality.
If you don't believe, don't hesitate and visit my temp homepage again.

http://www.volny.cz/martel

I put some test MP3s there so you can compare yourself. If you hear no difference between "spellbound-396r-b192-q0" and "spellbound-396r-b192-q3" then you should visit a doctor.

All the sample MP3s sound a bit different, the worst sounds the 396,q0 version.

"396r" in the filename means that it was produced using the 3.96 version, compiled by Rarewares, "b192" means that the cmdline switch -b192 was used and "q0"/"q3" means that the cmdline switch -q0/none was used.

A losslessly compressed sample would be much more helpful for Gabriel and others interested in reproducing the problem.

There is a wav file on that page. I suppose that is the original?

It is, indeed! If it wasn't i wouldn't have put an uncompressed file on web. Although it might look so, i ain't a complete maniac...

>dev0: You didn't have a look at my page before you posted, did you?

Some other thing needs to be discussed - our test equipment.
I have the SB Live! 1024 soundcard (about $30) and the Sennheiser HD 490 Live headphones (about $60). Both are pretty average pieces of equipment, but they are enough for MP3 tests (since they ain't audiophile).

I wrote this because i didn't want to end up arguing with John Doe that has a $10 soundcard and $5 headphones...

Perhaps, they could add a bracket to a member's account at this forum that would make it possible for others to view his/her equipment without asking. That would sure solve the "John Doe problem".

John Doe isn´t a problem as long as he has nice ABX results.

Well, forget about John Doe. Someone might feel offended...

On which -qx is the quality worsening?

quality 0 (highest precision)

Only when switching to q0? So using q1 everything is fine?

I don't know if EVERYTHING is OK, but it sounds much better when q = 3. Hear it yourself, the samples are on my web page (see previous posts).

I know it, you have already written it.
What I am asking is, as you are noticing a degradation between q3 and q0, from which q value are you noticing this degradation. (is it at q2, q1 or q0?)

Well, why don't you try out yourself? I thought you were the ones to hear the grass growing...

I would say that -q1 is "better" than -q0, but i ain't sure if -q3 is "better" than -q1.