To tell the truth, this sample is not all that new. I already used it to test Vorbis 1.0 and even posted it to the vorbis-dev mailing list once. However, the problem still exists.

About the sample: It's originally from the Farbrausch-Demo "The Product". This is a 64kb-demo/intro, so all the music is synthesized while playing the demo. Farbrausch later released a music disk called "Brullwurfel" (basically their synthesizer code, a few songs from their demos and a nice interface), also 64kb large, that included the track and had the ability to synthesize to a wav file.

At low q-levels, there is very audible noise added. When the q-value is chosen above a certain threshold (-q6 for older encoder, -q5 for the newer ones), this noise vanishes (almost) completely. This could be the point where lossless stereo coupling is enabled (at least that was the explanation I got at vorbis-dev). However, at -q6 even with AoTuV r1, there is still (subtle) audible hiss, most clearly audible at around 22 seconds I think. The problem is present with all versions of oggenc I tested (1.0.1, 1.1.2 and Lancer's 20060903).

You can download the sample here:
the_product.flac

Comments welcome.

where can I get it?!?

Here's a download link to the "music disk": http://www.scene.org/file.php?file=/demos/...ip&fileinfo

EDIT: It's track 2. (fr08main)
EDIT2: oh, and here's the link to Farbrausch's "break-through" demo .the .product: http://www.scene.org/file.php?file=/demos/...ip&fileinfo

Oops! I prepared and uploaded the sample, and then I forget to post a link!
Here you go:
the_product.flac

Sorry for that. I'll edit the OP and add the link there, too.

Your problem might be completely unrelated, but what your hearing is Noise Normalization if you are using less then a -q 3. If you have a pair of headphones on and are using -q 2 or below it's very noticeable.



If it's that bother some you can use the impulse_trigger_profile switch to see if it makes a difference or you can use it inconjunction with the impulse_noisetune. Just a thought. It could very well be the recording too.

No, the noise is still there at -q4 and almost vanishes at -q5



I think I'll try it this afternoon. I did this before (when testing Vorbis 1.0), but it didn't reduce the noise you can hear below -q5. maybe it can help with the noise that's still present above this q-setting...



I ABXed the encoded version at -q6 against the original, both with Vorbis 1.0.1 and Lancer's build. The difference is small and I would have missed it without an ABX tool, but I normally can't ABX Vorbis -q2 against the original with "natural" music.

Testing with different settings right now.
I just tried -q8 and am still able (though with difficulties) to ABX. Result was 13/17, the difference is really hard to get. There is a bit of noise around 22 seconds into the sample, near the center, which is slightly louder or more noticeable in the encoded version. Note that the average bitrate was 337kbps with this setting...

With -q6 and impulse-trigger-profile=6, I was unable to ABX. I still got more right than wrong but it wasn't significant. Bitrate with this setting is 296kbps.
Encoding was done with libvorbis 1.1.2 (OggDropXPD with Impulse_Trigger_Profile from rarewares)

Also, re-reading HotshotGG's post, just a small clarification. I don't post this because it personally bothers me much, but because I think it's useful for the devs to have some problematic samples around. So I thought I'd share this one.

the_product.flac (working link)

I thought I had checked that link
Stupid mistakes without end... thanks for correcting.

Yes, I'm fairly familiar with this sample cause I listen to that track at least once a week since the demo's release.
Their "fr-010 (art)" and "fr-minus-03.2 (farbomat)" soundtracks may also be good for noise/microattacks testing.