[LAD] Floating point Denormals: C++ and Waf

Fons Adriaensen fons at linuxaudio.org
Thu Aug 2 21:17:58 UTC 2012

On Thu, Aug 02, 2012 at 09:56:51PM +0100, Harry van Haaren wrote:
> How small is inaudible? Or better yet, how small is way beyond inaudible,
> but enough to eliminate any denormals?
> The other problem is locating exactly where the problem exists: I'm using
> Faust <http://faust.grame.fr/> to generate the reverb DSP code, so
> injecting DC offset would involve me tinkering with the reverb algorithm.
> I'd prefer leave it alone, but the outputs of the reverb can be DC-offseted
> easily. I don't think that would be enough... since the denormals could /
> would also exist internally in the reverb algo?

Yes they would, at the output it's too late.

Any IIR will have chain of delay elements, with both the
feedback and the output being a weighted sum of those.
Addin 1e-20f to the input of the delay chain will avoid
denormals and be inaudible.



A world of exhaustive, reliable metadata would be an utopia.
It's also a pipe-dream, founded on self-delusion, nerd hubris
and hysterically inflated market opportunities. (Cory Doctorow)

More information about the Linux-audio-dev mailing list