Sujet Mastering a -0.3 db ? pourquoi

Citation : Bearing in mind what’s just been said, it might seem logical to keep the signal
peaking a few dB below digital zero until all processing has been carried out.
After that, you can safely normalize the signal–or can you?
A related problem with peak clipping can arise when a normalized soundfile or
signal is converted to a new sampling rate. The reason has to do with the
sample-rate conversion process itself, and during sample rate reduction,
the signal is effectively being filtered; the available audio frequency range is
smaller at lower sampling rates. Such filtering can increase peak sound levels
in exactly the same way as attenuating equalizers can. But even when
increasing sampling rate, an increase of peak level can occur. This is because
the continuous-time audio waveform is represented in the digital domain only
by its values at the sampling instants. It is perfectly possible for the peak
value of the continuous-time audio waveform to occur at instants lying
between two sampling instants, and thus to be higher than the peak value at
any of the sampling instants. When changing the sampling rate, new sampling
instants are chosen for the continuous-time audio waveform, and these new
sampling instants may coincide with an increased peak lying between the
original sampling instants. This is especially likely to occur with signals with a
lot of high frequencies, since such signal waveforms change more rapidly
between the sampling instants.
Though artificially contrived signals can be created to really show up this
problem, in real life an attenuation of at least 0.3 dB or so prior to conversion
should provide adequate protection against clipping. You might expect
sample-rate-converter designers to account for the possibility by designing in
a small amount of attenuation, and the cheap ones generally do not.
But can you safely normalize a file that you know is at the final sample rate?
Unfortunately not, because many compact disc players (and some other
digital consumer equipment) use over-sampling digital-to-analog converters
(DACs) to produce the analog signal fed to the amplifier. Such over-sampling
converters involve a sampling rate conversion process which can (and does!)
cause audible peak clipping. Once again, some designers appear to have
overlooked this problem, although not as widely as they did in earlier DAC
designs.