I have begun to try to determine by ear, using Winamp and my personal retail CD's, the general volume level at which MP3Gain's scale and my original CD's sound about the same, at least as close as you can get with MP3Gain (which is, pretty much, close enough). The MP3's were made using Lame @ --alt-preset extreme. This is not a scientific test by any means, just a practical exercise.

Early results point to a level of about 92 instead of the default 89, but it's early yet. This has me wondering if a setting of 90.5, for example, might be closer to the general volume level of a CD (standards, classical, oldies, barbershop, opera - no hard rock or metal) than the default of 89. There doesn't seem to be a problem with clipping at 90.5, either.

Does anyone have experience or knowledge with this?

I don't get what you're asking. If the volume at 89 doesn't sound the same as your cd, then turn up the wav volume. Or do you mean "sound the same" in terms of some sound quality thing that you think is affected by the gain level, independent of volume? Concerning the literal level, 89 dB is somewhat arbitrary, but its functional basis is that most mp3's no longer have clipping at that gain level.

It's not completely arbitrary AFAIK... movies are designed to play at 89db as well.

The whole point of the default being 89dB to avoid clipping. Sure, it's possible to go higher without clipping for most music, but there are some songs out there that will clip even if the gain is 90.5dB. At 89dB there are very few.

I mp3gain my files at 92, simply because my mass majority do not clip there AND my portable doesn't produce a very loud signal. I'm sure that when I get a headphone amp, I will re-gain my mp3s to 89.

What I mean is the volume level apparent to the ear, that is to say, how loud the track sounds to your ear. Following was my procedure:

1. I ripped tracks from a good quality CD using EAC. I then gain normalized the tracks to a peak level of - 0.20 dB.
2. With the resulting Wave files, I created mp3's using Lame --alt-preset extreme.
3. I then analyzed the mp3's with MP3Gain and got those results.
4. Loaded the analyzed mp3's in Winamp.
5. Loaded the 'mother' tracks from the original CD (the track that EAC ripped from) into the same Winamp playlist window.
6. Then, with neutral settings (medium volume with Winamp (74%), medium Windows volume, no equalizers, etc., etc.) listened to a track with headphones, then immediately listened to its parent track from the CD, and so on, until the volume difference (if any) was evident.
7. Finally, using the Track Gain feature inside MP3Gain, adjusted the mp3 track as close as possible to the CD track, provided the adjustment didn't clip the mp3 track, so that the volume, gain, loudness, etc. sounded the same. Most tracks, after the Track Gain adjustments, sound identical in loudness to the CD's. The few that don't (because of the 1.5dB increments) are close enough to the CD volume that it probably doesn't matter (to me, anyway), since only the sharpest ears will notice.

This procedure results in a reasonable idea, it seems to me, of what gain level the original CD track is, when converted to the scale that MP3Gain uses.

So what I'm saying is, while the gain values for the CD tracks varied as you would expect, the majority of them, so far at least, seem to average somewhere around 92 as opposed to 89. This finding, based on a small sample, caused me to wonder whether a value like 90.5, used as a benchmark, might result in mp3's that sound closer to the same loudness as the original CD, without compromising quality via clipping. And I wondered what others thought.

Since I am dealing with a large library of files this interests me, and I wondered if anyone else may have had experience with this, or noticed the same thing. This would be useful to know at this stage. Interesting enough, and it's way too early to tell, for the most part only the tracks that the Track Analysis say need attenuation need adjustment in order to approximate the CD's. In other words if the analysis says to lower the level by -1.5, -3, or -4.5, etc., that usually results in about what the CD level is. But if the reading says 87 or 89, and suggests a gain increase of +1.5 or +3, etc., no adjustment is usually necessary to equal the loudness of the CD.

FWIW, I have no qualms with adjusting my mp3's to an average of 89 if that is the best thing to do in terms of quality and practicality. But if 90.5 or 90 or whatever, sounds more like the real thing without compromising anything, why not? That's all.

Sorry if I haven't been clear.

Hmm, I am not an expert but...

Are you by any chance just looking for a way to get your MP3:s to have the same loudness as the original CD? Then I think its suffiecient to skip the scaling in step 1 and just encode with lame ape, it doesn't change the volume. If you really want your MP3:s to sound as close to the original as possible you shouldn't do any normalizing or MP3gaining.

I don't think so. MP3gain is a practical implementation of replaygain.
Replaygain's goal is to achieve equal perceived loudness at playback. In other words: you set a playback volume you're comfortable with and you don't have to adjust the volume control at every next track or album. See the Replaygain document section
That this also helps to prevent clipping is great, but the 89dB "reference" standard has not much to do with that. (with todays (loud) music you could set it somewhere between 92-94 with hardly risk of added clipping during decoding).

I was using 92.5 for some time , it was a compromise to better match other sound sources on the computer. But now I'm using Musepack more, where replaygain has a fixed value of 89, I've moved to 90 for mp3gain.
--
Ge Someone

What you are trying to do is completely hopeless, in fact it's one of the reasons why ReplayGain was made.

CD's do not have the same loudness. They vary largely from CD to CD, and even more largely as time progresses. Older CD's are almost always less loud than newer CD's.

MP3Gain was made to set all CD's to the same loudness, and certainly not to try to approximate the loudness of the original CD, on the contrary even.

The 89dB was picked as a compromise between having headroom and being 'loud enough' so the difference between replaygained and nonreplaygained files is not too large in the typical case.

It will not be possible to get a single gain value that makes all your recordings come out like the original CD's. If that were possible, there would have been no need for ReplayGain. At best you can take a shot at a good average, which will vary depending on the average year the recording was made on.

IveyLeaguer,

You've been perfectly clear. I've read in another thread what you're currently trying to do - so I realise that you're taking files which you've modified (including peak normalisation), and then using mp3gain to equalise the loudness of them all, correct?

And what you're discusing in this thread is finding out what a typical loudness on a typical CD really is, so you can make all your tracks match that. Correct?


The answer is: depending on your choice of CDs for analysis, you could get almost any value. Looking at Frank Klemm's list, he has albums ranging from 85.5dB to 99.0dB in his demonstration replaygain output (I assumed all the figures he gave were relative to 89dB). There are individual tracks with a much wider range listed on the replaygain website.

If you take the 99dB album, and reduce the gain so it plays back at 85dB, it'll be a lot quieter - but that's all. If you take the 85.5dB album, and increase the gain so that it plays back at 99dB, it will almost certainly clip like mad and become an obviously distorted mess.


If you're using mp3gain to change the gains of the files themselves (rather than the mpc or ogg versions of replaygain, which store the values in tags) then it's a balance between getting all your files as loud as possible, but preventing any of them clipping. If you're sure that all of your files DO NOT clip at a certain value, then it's fine to use that value. 89dB is suggested, but you can use anything you like.

Note that, if you later find a file which clips if raised to (say) 92dB, then you can either let it clip, set it to 89dB instead (masking it a different loudness from the other files you have), or take all the other files down by 3dB, to make them 89dB too.

Finally, the mpc and ogg versions of replay gain assume either 83dB or 89dB as a reference loudness (I've lost track of which one they actually use - pathetic, I know - but replaygain isn't the only thing I've worked on, and I haven't kept up to date with developments). If you use 89dB in mp3gain, then mp3s, mpcs, and oggs will all match in loudness using default replaygain processing in those formats. Because ogg and mpc replaygain uses player processing, you can change the default values in the ogg and mpc plug-ins in winamp (say) so that replaygained oggs and mpcs do match your 92dB mp3gained mp3s.


It's up to you, but unless you find 89dB too quiet, you might as well use it!

Cheers,
David.

P.S. - Jebus - film soundtracks are mastered using 83dB calibrated equipment, with 18dB headroom above this. When people use 89dB instead with replaygain, it's not because people are playing their music back louder than a movie - it's because the audio equipment is set to a lower volume than a cinema system, so the digital audio in the file has to be louder - twice as loud = 6dB higher = 89dB. This leaves only 12dB headroom - more than enough for most modern pop recordings, but not enough for the loudness difference between a whisper and an explosion - unless you make the whisper a bit louder and the explosion a bit quieter! You'll find just such an option to reduce the dynamic range on some DVD players and AV receivers, to make the film soundtrack easier to hear at home.

EDIT: Thank you Garf, you beat me to it!
GeSomeone (who also beat me to it!), you'll see that I think that 89dB is a balance between loudness and clipping.

Posted by 2Bdecided on Apr 1 2003 - 04:13 AM
Posted by Garf on Apr 1 2003 - 03:48 AM
Boy, isn't that the truth. I've talked about this also in older threads. One extreme, but real, example: I had 2 copies of one recording (The Lettermen) from different 'first quality' CD's, produced by the most well known studio of all time. When examined in graphic Waveform, one copy was good, well defined with a peak gain of about - 0.4dB. The other appeared as a straight line, approximately the thickness of a shoelace. No kidding.

Posted by Garf on Apr 1 2003 - 03:48 AM
Good point, yes, and other factors as mentioned above. So I should take a shot at 90.5, say, and try it, but the result should be kept consistent - that is, determine a number for MP3Gain and use it for everything, adjusting individual tracks relative to that number if I want to be more precise, right?

Posted by 2Bdecided on Apr 1 2003 - 04:13 AM
Well, that is really good information because once my Waves are losslessly archived on disk, I'll be going to MPC & Ogg, then later to MPEG-4 or whatever the community comes up with. In fact, though I'm quality nut, I plan to archive with Lossless on discs, then burn CD's from MPC for now, especially standards, oldies, choral, etc.. Actually, Dibrom's --alt-preset extreme works well for those categories. Can I hear a difference between an oldie burned to CD from mp3 vs. ape? Sure, most of the time, in a quiet house with my headphones. For high quality classical, I'll still burn from Lossless. But the way it's going, that might not be necessary for long, though I suspect the need for ReplayGain will go on.

Posted by 2Bdecided on Apr 1 2003 - 04:13 AM
I think I'm gonna shoot for 90.5 because that's just me, but I'll not be surprised if I end up at 89. It's looks like a win/win.

So, God Bless You All and Thank You.

2Bdecided:

I was under the impression (from the replaygain website actually) that 89db was chosen because it matches the target volume for films, the justification being "why not choose a value that is already a standard in parts of the industry?".

Seems to me though, as I'm looking through it right now, that the replaygain folks recommend 83db like you stated, so my assumption that it is 89db is flawed. the question is then, "why does mp3gain use 89db and not 83db?"

Some of it is explained here: http://www.digido.com/index/pmodule_id=11/...der_page_id=59/.

That was a touch over my head, unfortunately. Did they say that theatrical releases are calibrated at 83db and home releases are calibrated at 6db higher (89db)?

If not, please clarify

edit:

Nevermind, it has occured to me that the reason movies are at 83db is so really big stuff (explosions etc.) have lots of headroom, while music can play at 89db because the dynamic ranges arent as large.

Musings of a MP3Gain newbie.

Well, FWIW (I know), three categories have been tested (by ear), and exactly none of the results were what I expected. The idea, of course, is for the volume output to sound pretty much the same, regardless of the type of recording. The early results:

Oldies - I thought 92 might be a good level, but a couple of samples later I started by using Track Gain to 89. Through listening, I started over at 90 which seemed to be a good level with no clipping. Because the files were normalised, most of these were in the mid nineties, some higher. At 89, these seemed just a a little low. So, 90 worked well with the usual exceptions (10% or so), which have to be individually Track Gained up or down.

Barbershop Choruses - 92, because of a capella, and some very old recordings. Most of these had to be adjusted one by one.

Classical - The real surprise. I was pretty sure these recordings would require a higher benchmark number than the default of 89, and maybe higher than the 92 Barbershop required to hang in there with the Oldies.
Wrong again. The best level so far? You guessed it - 89.

BTW, the article by Bob Katz, linked above by CiTay, was very informative.