-

You must separate different kinds of clipping.

Many modern over-compressed recordings are said to contain clipping, but that is not a technically valid claim. These recordings may be over-compressed and sound bad, but the maximum peaks are always 1.00 (aka 0 dBfs, aka full-scale) or below.

A lossless source file does not produce clipped peaks on decoding if the decoder works properly. You can make a lossless file (or any other audio file) to clip by using inappropriate playback DSP settings that create clipped peaks. This is easily audible if the used bad DSP settings are strong enough. (For example, you can apply a +20 dB replay gain correction without a clipping prevention or use strong "+ EQ corrections" without reducing the overall gain respectively)

Lossy files can produce so called decoder clipping, i.e. the maximum peak values can be over 1.00. This is what MPGain reports as clipping. It depends on the decoder and playback software how this situation is handled. I have not seen a single reliable report of the audibility of this kind of peaking. I have tried to ABX the difference, but because these clipped peaks usually have a duration of less than one ms I have not been able to make any difference between clip prevented and clipped samples.

Edit: typo

It's ABXed here:

http://ff123.net/norm.html

The (now long gone) forum discussion that prompted ff123 to do this test was from someone who could easily hear this clipping. This was long before there was mp3gain to tell him there was a problem - his ears told him something was wrong, which led to this investigation.

On the samples where bAdDuDeX could easily hear a problem, I could hear nothing. When the MAD mp3 decoder added even more clipping (it was buggy back then - it's fine now), I couldn't hear that either.

Cheers,
David.

Thanks, that's an interesting report. Somehow I have missed that page.

ff123's test setup is quite similar with what I used except that I did not adjust the source files before encoding. ff123 compared encodings that were made from two different source files.

I used only one set of encoded files and decoded my two test sample sets like this:

- one set with foobar's replay gain adjustment (thus preserved the wave format of the +1.0 peaks). I used the minimal amount of adjustment that prevents clipping.

- another set without a clip preventation (thus truncated the peaks). After decoding I applied the same amount of negative gain that was used with the replay gain adjusted set. (I used Wavelab for this.)

I'll try ff123's samples and report later.

That heavy metal sample is horrendous!!! Much easier to pick up the clipping with synths or classic music.

But the question is... since you guys didn't hear the difference, that is, didn't hear any clipping at all... wouldn't it be because it requires that you listen to the sample with the maxed out volume of your equipment? (see question 4 of my first post).

You really don't need to worry about clipping when using modern encoders like LAME...

There are very very few samples that have reported clipping using standard presets of v2 or equivalent.
JXL

This is strange... with Musepack, that is a good encoder, nearly all my CD clip (replaygain album peak value > 1).

In one of them, the clipping is even audible as a sharp click.

pio2001, could you post that sample!?

I can't find it anymore !
It was at the end of Transwave - Helium. I just listened to four tracks of this album where it should be without hearing it... I'm listening to Phototropic, just in case.

- posted in wrong thread -