Replay Gain - A Proposed Standard

Reducing the Gain adds noise!


7/12/2001, 9:52 pm

Matt wrote the following on the Monkey's Audio Message Board:

Well, keep in mind that you can only store amounts to turn tracks down, not up. Peak level normalisation (already in MAC), provides the most you can safely turn a song up without clipping, but this obviously doesn't even begin to solve the problem you're working on.

So, you'll need to actually store something like -12dB for the loud songs, but this is essentially the same as throwing away data. (thus introducing noise)


Agreed. But let's assume people really want to solve this problem. From the thousands of posts I've seen on various message boards requesting this very feature, it's clear that they do.

To maintain full dynamic range, the ideal solution is to feed the Replay Level value out of your PC, to your volume control. Obviously this requires dedicated hardware, and few people are going to do this, but it would be possible for those who demand highest quality to put an end to stupid fluctuations in level in this manner. No compromise. No downside. It'll appeal to Monkey's Audio users!

The (compromised?) solution we can offer now, to everyone, is to do what you suggest above - just turn the loud tracks down in the digital domain. With the standard I've proposed, it drops the level of a hard limited pop track by about 10 dB. There are two reasons why I don't think this is a serious problem:

  1. Such a track already has a highly compressed dynamic range. Most of the energy was within -3dB of the peak. Now it's within -13dB of the (system) peak. Still way way way above the noise floor of even the worst soundcards.
  2. Dropping 10dB is equivalent to losing 1.5-bits of resolution. The input data is likely to be 16-bit (with a CD source). You, and many others who care about sound quality, own 24-bit sound cards. The dynamic range is about 110dB. The noise floor of the original recording, even after being dropped by 10dB, is still the limiting factor - you're not going to loose anything in a 24-bit replay system.

I realise the absolute highest fidelity possible when feeding a 16-bit source to a 24-bit sound card is to send it as-is, to occupy the top 16-bits. So whatever else you do is a compromise. Well - if you feel like that, go and build the hardware I've described to alter your volume control according to the Replay Level!

But in reality, a consistent replay level (an advantage which EVERYONE can appreciate) will outweigh the problems of digitally scaling the data (which is a problem that 0.01% of listeners are going to notice).