The Badvilbel sample:

http://www.ff123.net/samples/badvilbel.flac

The hybrids are also affected during the silent parts. Again no 'artifacts' as we expect , but a boost in background hiss. The original is also hissing.

Wavpack 4.2:

5.2-6.5 secs

-hb256 : very easy to hear and abx 8/8

-hb320 : Better but obvious 8/8

-hb450 : Threshold: I have to raise to raise the volume above normal and then it there **must be carefull of the noisey parts !!** 8/8

-hb500: Seems to resolve, I am risking equipment damage and have no motivation to try abx again at this bitrate..

Using -hx makes it better, though a slight hiss is still there up to around 450k.

Listening to this sample naturaly I can't hear anything wrong at 400k +, but only when I abx the difference is heard. I only stumbled on it through abxing out of boredom.


Optimfrog Dualstream:

Quality 0 (260k): Easy 8/8 but sounds more subdued than wavpack

Quality 3 (360k): Sounds quite good, but there is a subtle hiss 7/8. Dualstream seems to inflate bitrate more than WV possibly due to its quality mode ?

Quality 4 & 5: (400-460k) Can't abx it even with extreme volume levels!

10.6 seconds is also abxed up to quality 3 or 4 (worse than wavpack !?). At quality 5 I gave up on 5/5. I am not sure if I hear anything worthwhile even on really high volume and I am scared of equipment damage.


I searched for other potential similar samples (quite bits) but I cannot find anything even close to this signal. It seems to have a really weird loud freq spike that pushes background fuzz upwards and the hybrids cannot mask the noise as they normaly do ? This must be a worse case scenario for hybrids, but it reality I would have never heard a thing during normal listening.

Thanks for providing this sample; I'll add it to my collection...

The problem here (as you guessed) is the high-level noise over 15 kHz. The "high" and "extra" modes help somewhat because there's a pretty significant peak just over 18 kHz that can be decorrelated out, but the rest is fairly broadband.

The only real solution for samples like this will come when I implement the psychoacoustic-based "quality" mode (which will simply crank up the bitrate in those sections). Fortunately artificial samples like this are rare.

And, yes, I wouldn't play this loud on real speakers because it could easily fry tweeters!