Am 18.06.2018 um 08:13 schrieb Felix Eichas:
> There's also a paper regarding power complementary crossfade curves.
> Maybe a bit scientific but still worth a read:
Interesting paper, I did not expect that this issue has been analyzed in
such a detailled way.
Anyway, there are some issues:
The mathematial power of a signal is related to it's spectrum, and if we
cross face two signals with different spectrum, than we have to make up
our mind which frequencies we want to focus at. Mathematically - and
this is done in the paper - it is easy to meassure all frequencies'
power and simply adjust the levels that way that they match - according
to the definition of power, which is related to the period as you know.
Well, this is not the solution!
The reason is that - depending on the particular application, individual
frequencies have a different "importance" in the app. This is the case
e.g. with radar sweeps, refelection triggering and similar things.
For us, here, dealing with audio, we have to take the hearing curves
into account, meaning, that at a specific loudness level, the
frequencies have a different impact, so simple level orientated fading
leads to wrong results. The here problem is, that some loud parts of the
music do create some kind of "mask effects" in the ear, so this
frequencies do not appear in the experienced power.
As a consequence of that, also the speed of fading (a flat or a more
steep curve) also has a significant impact on the loudness, we "feel".
Also for short cross fades, some frequecies hardly run into the
mathematical equation so also the algorithmic way is strongly depending
on the fading period and causes different results.
I typically have that problem, when putting together several takes in
orchestral recordings. The level meter is no help during this decision.
Instead listening is the only way to do that correctly.
With piano recordings I remember situations, where - due to the
complexity of the sound - it was nearly impossible to fade that 100%
because either the bass was to high or the disctant would have been.
So mixing is always compromise, because some musical notes do work as
accents in th flow and a mathmematical algorithms hardly can judge this.
The result of that is, that for example the level of a subsequent part
might already have to be changed, just because the flatness of the
fading curve is changed, which in theory should not be the case, when
regarding the signal power.
My option to this issue:
Signal Power is not equivalent to audio power and this again is not the
same as expericenced loudness and this again is not the same as musical
loudness impression in the a contex of a track. These are 4 "different
shoes" , as we say in germany.
dupswapdrop: music-dsp mailing list