[LAD] Floating point processing and high dynamic range audio

Paul Davis paul at linuxaudiosystems.com
Wed Jul 21 18:05:44 UTC 2010


On Wed, Jul 21, 2010 at 2:01 PM,  <fons at kokkinizita.net> wrote:
> On Wed, Jul 21, 2010 at 01:26:40PM -0400, Paul Davis wrote:
>
>> floating point -1.0 == minimum value of an integer sample == -inf dBFS
>> floating point 1.0  == maximum value of an integer sample == 0dBFS
>
> ??? At least the first of these makes no sense at all,
> and the second applies if the dB value indicates a digital
> peak level. More accurately: 0dB digital peak corresponds
> to a sample value of + or - 1.

yeah, i got the first one wrong. mea culpa. but i think you need to
avoid using "dB" in this context, and be clear about using dBFS.
restated:

floating point value of -1.0 == minimum value of an integer sample
floating point value of 1.0 == maximum value of an integer sample
floating point value of 0.0 == -inf dBFS
floating point value of +/- 1.0 == 0 dBFS

> An audio signal represents pressure variation as a function of
> time. Multiplying it by two will give 2 times the pressure,
> and 4 times the power. The subjective result is another matter.

i was trying to be precise enough to include nonlinearities in the
analog path. perhaps unnecessary.



More information about the Linux-audio-dev mailing list