Replay Gain - A Proposed Standard

Peak Normalisation

Why can't we just (peak) normalise the tracks?

A huge red herring when trying to make all tracks sound as loud as each other is peak normalisation. In this process, the level of each track is adjusted so that the loudest moment (technically the largest sample value) is adjusted to peak at the largest allowed value by turning up the gain the required amount. By making all the tracks peak at the same level, you might expect that they would all sound equally loud.

Unfortunately the peak level is a very poor indicator of perceived loudness. There are two reasons for this. Firstly, the human ear responds to average energy over time, rather than instantaneously short peaks. Secondly, many records are mastered via compressors and limiters, which aim to raise the average level of the sound as much as possible, whilst still squashing the peaks to (just) within the limits of the system. An uncompressed symphony can sound less than one quarter as loud as a compressed pop music track, even though the peak level may be the same. Even different pop records use different levels of compression, making the peak level useless for determining the perceived loudness of a track.