On or about 8 Jun 2001, andy at theasis.com is alleged to have said:
> > No scaling (I wanted to add utilities to do that but we did not need
them
> for
> > our project back then).  Just a linear slop to fade or cross-fade from
one
> > clip to the next.  (What curve is best for a fade--an S curve that is
> smooth
> > and flat at each end?)
> 
> What's the scale?
>  
> Since dB is defined in terms of Psychophysics, and it's a log (ratio)
> scale, I think you want something akin to linear on a log scale. That
> means linear in terms of dB, but it's not clear to me what physical
> property you're actually attenuating (power? voltage?) 

Whoa -- hold up, Hoss!  It's way easier than that.

When you are mixing, i.e., adding signals, you have a maximum value for
each sample.  So, generally, you scale back each input source to give it
room to add them together.  Typically, you reduce each signal by 6dB, so
that if you ever have a point where both signals are at full maximum value,
the resulting sum will not overload the system.

Now decibels are not anything as far as Psychophysics -- whatever that is!
;) -- decibels are simply one tenth of a bel, which is a unit of relative
change.  They are very handy for electrical, acoustical, and a couple of
other log scale measurements, but you could just as easily say that Joe who
makes $20k/year earns 6dB$ less than Sue who makes $40k/year.  You have to
specify the units, which in this case are understood to be dBFS (decibels
w.r.t. Full Scale -- whatever $MAXVALUE is for your wordlength.)

Does that make it make more sense?  

Phil