As I was reading about replaygain here: http://wiki.hydrogenaudio.org/index.php?title=Replaygain , I was wondering: relative to what level should this 89 dB be interpreted?

That's a sound pressure level (SPL) value.

You can read more about Replaygain at replaygain.org. If you search the web for the referenced SMPTE RP 200 standard, you'll find more information about the topic in this article by Bob Katz, for instance.

Erm... Perhaps I read the question a bit differently, but I don't think it's an SPL level, at least not directly.

Neither sound nor pressure exists in any encoding. It is in the interpretation/playback which turns it into sound. If something comes back as loud, you can turn down the volume knob, which in effect tells the system to not play it as loudly. I'm quite certain if you play something that's actually 89dBA on an SPL meter, you'll find it uncomfortably loud.

So a better question would be, how do you interpret volume based directly from an encoding? I imagine the answer has something to do with how one interprets samples.

Yes, actually my problem was that the sound pressure level (according to Wikipedia, the reference value for this is 20 micropascals) is a quantity describing "real" sound, not the representation inside the computer. So a music player cannot set the physical volume directly because it can still be adjusted by the user on the output hardware. But I found this on the replaygain.org page suggested by smack:
http://replaygain.hydrogenaudio.org/calibration.html

Does this mean that music players supporting replaygain process the audio data so that it will be 20 dB quieter (on the average) than the loudest digitally representable sound? (Or maybe just 14 dB because this article uses 83 dB SPL instead of 89 dB?)

Please correct me if I don't understand these things correctly.


EDIT:
One more question: If I read the article correctly, it is suggested that users set the volume knob so that replaygain processed files will sound 89 dB SPL loud on average.
But as sthayashi already stated, a 89 dB sound seems to be quite loud, actually Wikipedia says that long term exposure to it causes hearing damage.
So I guess I still do not understand it correctly ...

And why does replaygain.org talk about 83 dB while other sources mention 89 dB?

If you actually get 89dB of SPL at your speakers depends on the setting of your volume knob. As far as I know, replaygain decodes the samples and calculates the average loudness based on a model of the human hearing (something like dBa, I suppose). The argument about 89dB being too loud may be true, but nowadays, CDs are mastered very close to their maximum dynamic range, which is around 96dB. You wouldn't want to kind of volume in your living room, either...
But as long as enough of the dynamic range is being used, everything is fine isn't it? The point of the whole thing being that you don't have to reach for the volume knob on a new track.

Yes, this is exactly what brought up the question in my mind: the volume depends on the knob setting, so what does it mean that the loudness is normalized to 89 dB SPL?; I mean what does it mean from the point of view of the representation of audio data inside the computer? After all the 89 dB SPL is a measure of real (physical) sound, with a reference point of 20 micropascals, and it cannot describe digitally represented amplitude levels unless we also choose a reference point for them (decibels express a ratio, not an absolute value).

Doesn't the volume knob on your preamp only serve to attenuate the signal? So theoretically at 100% volume with a powerful enough amp, you would get 89dB SPL? I know that if I connect my pc lineout directly to my power amp it's unbearably loud unless I turn foobar's volume right down.

I can't give you a hint on the db in digital media, but where the original 83db come from is explained in Bob Katz's article that is linked in the second post.

Ok, you're right. To rephrase the question: how is a certain digital amplitude related to the (probably A-weighted) dB SPL in the physical domain?

The first thing that comes to my mind is a direct mapping fom dB SPL to dB signal/noise. But that would mean that a signal has to have a resolution of at least 15bit (6,02 * number of bits) to even reach the reference level of 89dB, and the scheme wouldn't work for lower resolutions. But then, it's late and I'm probably not in a top mental state, so I better stop here and leave the field to the real experts (god knows i'm not one of them).

By SMPTE RP 200, which says that pink noise at -20dB (with respect to digital full scale) should be reproduced at 83dB SPL (as measured at the position of the listener). It's just a guide for how to set the volume control. It was created for the cinema.

I think the ReplayGain website, Bob Katz's web site, and maybe even the ReplayGain Wikipedia article cover this.


Why 89dB? Because all the implementations just add 6dB to the target before storing the value. In the original proposal, the players were expected to do that, but this has changed and the adjustment needed to hit 83+6dB is the standard stored value now, so players don't add anything, just implement the adjustment as stored. Some people with a lot of classical music find they need to get their player to subtract 6dB to get back to the original standard, since shooting for 89dB clips some highly dynamic classical recordings.

89dB SPL is very loud. 83dB SPL is loud. Even 77dB is loud. I wouldn't listen to anything on a "calibrated" system if I had the choice, though some people do.

Cheers,
David.