Hey guys, I'm new here.

I've recently discovered MP3Gain, and I'm in the process of normalizing (average album output of) my entire music collection.

I have analyzed it all (with the default target normal volume of 89) and I've noticed that for the great majority this will be reducing the volume with a gain anywhere from -12 to -1. I guess there might not be anything wrong with that -- I put a normalized test album on my iPod and it sounded fine. But this is a really massive music collection (130GB) so I'd really like to get it right the first time. Any suggestions as to what target normal volume I should be using?

Also what does clipping mean? Almost all of my tracks appear in red and have a "Y" for clipping. How is this different to clip(Track) -- almost no tracks have this.

Any suggestions would be appreciated.

Cheers,
sportsfan

I suggest leaving it at the default value.

Clipping means that the volume of a song has been raised by a certain amount, and in doing so the peaks of the song have hit 0 dB.

This will only happen very rarely on songs that are very quiet. If you increase the value above 89 dB you will increase the chance of introducing clipping to the songs. If this happens a bit it is doubtful you will notice it, but if it happens a lot the files will sound distorted.

You can use the option to avoid clipping, which means if the file needs a 3 dB increase to get to 89 dB, but the file will clip, the file will only be increased by 1.5 dB to avoid clipping. The trouble is that means that file will be significantly lower volume that all your 89 dB files.

So I suggest leaving the gain at the the default 89 dB level. But leaving the avoid clipping option turned off. If you get files that clip badly you will be able to hear it, and then can fix those files later by simply removing the gain and reapplying it with the no clip option selected.

I recently MP3gained 15,000 MP3s, and have so far only noticed bad clipping on 2 albums which I simply fixed by removing and redoing the MP3gain.

Hmmm okay.

It appears as if a great deal of the songs will be clipped. I'm willing to give this a try but before I do, let me just confirm I've got this right:

If I tell MP3Gain to "remove tags from files" that will completely put my files back to their original state... correct?

This would mean that you are listening to good music with plenty of dynamics. Most current music is heavily compressed and will be adjusted with values between -5 and up to -7 db or more. As you reduced the volume, clipping can't occur in these cases. It is only when the audio requires positive adjustments that there is a chance of clipping. I always replaygain preventing clipping.



Shortly: yes.

There are two ways of replaygaining for mp3 files, both of which are lossless (i.e. maintaining all audio data and reversible). One way is where tags are added. Replaygain aware players then use the tag to adjust the volume of the mp3 during playback. The actual mp3 data is not touched.

The other mp3 -only way is where the scaling factor of each audio frame is changed. Here, the actual audio data is changed. Advantage compared to method 1: compatibility, i.e. the volume correction works on each player that reads mp3. Drawback: this can only be done in steps of 1.5 dB. Practically not an issue though.

For the latter form of replaygaining, mp3tag adds "undo" tags. Using these tags, mp3gain can undo the change to the audio data, and you have exactly the original audio data back. In the first case you just remove the tags (or let your player ignore them).

Actually nvm, I've tested it out and discovered that I need the tags to restore the tracks to their original state.

Well, technically the mp3 can clip with no adjustment, due to the nature of lossy encoding, and could theoretically clip even with a small negative adjustment, but probably only with a very small negative adjustment. For all practical purposes your statement is correct.

Thanks for the info guys. And thanks for clearing that up chromium. Gotta love those cross-posts .

I've album-normalized my collection, and I'm not seeing any problems whatsoever. I'm definitely liking the reduced need to adjust the volume.

Sportsfan, if you want to hear some clipping in action there's a good (or rather, bad) example here (headphones will help).Oh, the joys of "remastering" back-catalogues.

C.

Although it might make a good case against dynamic range compression, I don't think that example has anything to do with this discussion.

EDIT: Not only did any of the examples show any obvious clipping, the lossless ones appear to be transcoded from a lossy source.

EDIT2: Going back to the lossless sources for this example there is in fact obvious audible distortion as as the peaks do get squared-off (minus the ringing). In addition, these peaks do come very close to full-scale in the lossless source. If one were to relate this phenomenon back to this discussion it would be that the lossy conversion of this type of thing would likely require an attenuation of the data in order not to be limited at full-scale upon decoding. The problem being that the lossy encode won't contain all of the information needed to keep the signal from going beyond this range.

If one were to convert one of these remastered titles to lossy, use of MP3 gain would not be able to remove the audible distortion.

Now here's a good illustration of how converting to a lossy format can cause clipping...
http://www.hydrogenaudio.org/forums/index....mp;#entry428436

Greynol, sorry if I was OT. My point was simply to provide an example of what clipping sounds like. Seems a possible misunderstanding of mine left me OT when I thought I was on topic.

Are you saying that the static crackling distortion on the sample is not caused by clipping?

I've just listened to my remastered 1999 release and I can guarantee that the distortion is on the CD (i.e. is not purely a result of lossy encoding/decoding and matches the sample), and is not on the non-remastered version. I thought that "squared-off" peaks and clipping ultimately resulted in the same effect, just that one is the recorded outcome of limiting and the other is what happens during decoding (as your graphic example highlights very nicely) - but both ways the peaks get "squared off", don't they? I've got a feeling I'm wrong. I'd be grateful if you could put me right on this.

ps. Thanks for the link; nice explanation.

C.

Oh, ok. I think the misunderstanding was mine then.

Well I'd say it was caused by the very poor use of limiting but I don't have a problem saying the crackling was caused by clipping. If you zoom in (an important step in visually identifying clipping which that website doesn't mention), you will see that the peaks actually do not run out of headroom (nor do they hold the exact same amplitude) but clearly it doesn't prevent the distortion. I guess this could ultimately boil down to definitions when tying together what you see versus what you hear.

Right. See my second edit, I did exactly the same thing.

Certainly. I'm not so sure that the effect is as dramatic in lossy decoding without a level adjustment, but I will attempt to do some testing with a highly compressed lossless source that doesn't sound distorted by converting it to mp3 and back to PCM again.

Thanks for the clarification.

C.

I made some clips using Lame 3.97 -V3 --vbr-new for comparison. One has clipping as a result of the decoding process while the other doesn't. I haven't used the greatest hardware to make a comparison but I don't think I can tell the difference between the two. Maybe you'll have better success...

Click to view attachment Click to view attachment

EDIT: It seems as though spaces, parenthesis and dashes have been replaced by underscores in the file names.

The files were originally titled, "clip (mp3, -3dB applied to global gain).flac" and "clip (mp3, -3dB applied after decompression).flac" prior to uploading.

Hey carpman, greynol

I listened to the flacs from the site carpman linked to, and to be honest I'm not 100% sure I could tell the difference between clipped and non-clipped.

I feel as if there was a slight distortion at the start of the clipped file, but I didn't notice anything throughout the rest of the file.

I'll try out these new ones you've uploaded greynol.

Cheers,
sportsfan

Illustrates your point nicely. Square peaks, but not perceptable (at least to me). Shows how clumsy the Doors remastering was.

@ Sportsfan -- did you listen to the samples with headphones?

C.

This doesn't mean that there aren't examples out there that are more audible, of course.

Going back to the target volume with MP3Gain, I use 92, but I listen to mostly rock (the sample I posted being no exception). With older material that has not undergone the heavy dynamic range compression commonly found today, I may run into a few more instances of clipping that I may not have had I stuck with 89, but it isn't anything that I ever notice while listening.

If you listen to other music that is more dynamic in nature then 92 may result in clipping upon decompression that is more noticeable, the same may even be true using 89. The original reference was 83 and with some music you may have to go that low to prevent audible clipping.

Ultimately I think what we hear matters far more than what we see.

I concure with what everyone providing advice has said so far, ...Unless of course you have a monitoring environment that is calibrated for reference-level audio. And in that case, 83 db is the choice - assuming that replaygain is accurately adjusting your music to the right "average" loudness. The "average" is the part that's really left up to perception between listeners and environments, but within replaygain itself you're basically stuck with it's automated process, besides the db level setting. So you may need plus or minus a few decibels depending on how replaygain is interpreting the type of audio you're running through it.

But for a lot of portable players, and boomboxes, table-tops, etc... 83db is not an option if you can't turn up the volume knob far enough to your liking.

Sorry, I don't quite get what exactly is clipping. I thought it occurs when parts of the peaks have been chopped off and have become straight lines when viewed zoomed in? So how does whether a lossy or loseless format affect this?

Yes, that's why I'm confused, from that link, I read "So the clipping only happens when you want to decode (there is no clipping inside the mp3 file)." But with most modern CDs nowadays, won't the clipping already occur in the raw data on the CD itself? So when you encode to mp3, it will have clipping as well?

As I understand it, you could in theory have both happening at the same time:
i.e. you could have a clipped source (due to poor mastering) and then get clipping (in the same or in a different section) due to decoding from a lossy encode of that source (as described by the link example). So clipping can be set in stone at the "production end" (where there's nothing you can do about it), or it can be a result of decoding at the user end (where MP3Gain and other tools come in handy).

I think what may be confusing is the "clipping only happens" bit. But, as I understand it, this is only in regard to decoding. They are not talking here about clipping at the "production end" i.e. on the "raw data".

That's my understanding. If I'm wrong, someone please correct me.
Thanks
C.

You're right Carpman.

Instead of saying "Now here's a good illustration of how converting to a lossy format can cause clipping" I probably should have said "Now here's a good illustration of how converting from a lossy format can cause clipping". The data inside the lossy file illustrated by the graphical example is not clipped, but the output will be if the gain is not reduced.

I see, so you're saying that decoding a lossy can create additional clippings. Thanks for helping me clear up my confusion!

Exactly.
Yes, it's been an interesting thread, cleared up a few things for me too.
C.

The second method is used by MP3Gain by default. However I can't seem to tell MP3Gain to use the first method (and act like foobar). How to switch from the frame-changing method to the tag-only method?

Anyone?