A commonly performed task is to bring the level of a recording to some standard volume level. This can be performed through the Normalize command. A problem with this is the notion of ‘level’, as there are many different ways of measuring a sound’s level. The Normalize function supports two notions of sound level. The maximal sample value simply looks for the sample that has the largest absolute value and then amplifies the whole sound in such a way that this sample is brought to the reference level. This has the advantage that if the level is chosen to be smaller than 0dB, it is guaranteed that no clipping will occur. The disadvantage is that a short outburst of unusual loudness or a crackle in the sound will dominate the picture, so that the achieved effect is not always what was intended.
The maximal RMS power setting tries to compensate for this by taking average levels over relatively short time intervals (100ms) and choosing the largest of these average levels as the reference. This has the advantage of being usually closer to the desired effect. However, it does not guarantee that no clipping will occur. Normalizing the RMS power to around –10dB or –15dB is quite safe in most situations.
Note that the Normalize function adjusts the global volume
of the sound, but does not compensate for variations of volume within
a recording. If you wish to obtain several songs sounding very
homogeneous throughout (as required for example for background music
in a store), you may want to apply dynamical compression before
normalizing. One sequence of steps that works well for most situations
is to first apply Apple’s AUDynamicsProcessor Audio Unit
with the hard
factory preset and then to normalize the RMS
power to 25%.
