Clip is here, taken from 4:49 into the song (remastered and expanded version of the album)

I tested with 1.14, quality 5 --xlevel. There's a bit of a difference as the cymbal hit fades out. The high end sounds a bit rougher on the compressed version. I was able to ABX it 14/16. I was also able to ABX with the newest Nero AAC on the standard preset (14/16) and the latest compaact! beta at q5 (15/16) with the same sort of artifact.

This track is clipped by MPC encoding. Try starting with this lowered version

Pio, please could you provide an explanation of how this sample you've provided has been processed?

Very basicly, I decreased the volume, and reencoded it. I saw that there was no more sample at 0.00 db in the decoded file.
The process is a simple 16 bits volume decrease (-1 db) without dither. You can produce a better sample with 24 bits processing + dither.

You seem to be right. I went 0/4 on your clip and gave up.

I thought --xlevel was supposed to prevent the encoder from clipping, why is it that this track still gets clipped?

Ouch! Trying to ABX such short samples gets on my nerves very quickly, could you please provide a slightly longer sample next time? Five seconds would probably be sufficient, however the convention seems to be 10 seconds or more, but obviously less than 20 seconds.

My apologies. It looks like Pio's already found the problem, but I've uploaded a slightly longer clip if you'd still like to play with it.

Quite impressive ! There were only 4 isolated clipped samples, and SoundForge didn't even report any clipping (since one isolated sample is not considered as a clipping).
On the lowered version, the highest sample is at -0.27 db. Thus it was clipped by 0.73 db.
A 0.918 scale is needed to prevent clipping.


Here are the three first clipped samples.

--xlevel prevent the encoder from getting internal clipping. The output from the decoder (which is the sum of several internal values*) may still clipp. These two types of clipping is not directly related to each other.

*Maybe it is not only a sum...?

This is totally off-topic, but strangely when I saw this thread at the HA main page, I was actually listening to "And you and I" the song ahead of "Siberian Khatru".

I was listening to Close to the Edge, specially my favorite "movement": II.- Total Mass Retain.

BTW, has anyone tried encoding this sample but from the older, WB version (the not-expanded one)?
It's the one I have, but I don' have access to the CD now, only to my MP3's (and won't be able to access the CD in quite a while).
Somehow I suspect that if we are talking a volume issue, the older version might "behave" better.

Ah, I see now.

Still, it seems rather unlikely I'd pick up on single clipped samples. My ears aren't that good
I wonder if something else might be at work here, tho I was unable to ABX Pio's lowered sample. Very strange.

The clipped samples are like clicks. Since the most clipped one was decreased by 0.73 db, it is like if a click had been added to the music. Its level is 1-0.918=0.082 times the full scale, thus
20 log 0.082 = -22 db. Which can be audible, depending on the conditions.

And it is difficult to know how four close clicks sound as a whole. Certainly like a "scratch"...

So it is well possible that what you ABxed was actually the four clicks.