Re: Development Summary (5.8..ba24f287ca)

Fons Adriaensen <[email protected]> writes:

> One more feature: have a look at
>
>   <http://kokkinizita.linuxaudio.org/linuxaudio/test3.png>.
>
> 'Bus 1' has a 1 kHz sinewave at -10 dBFS as input. It has an 'a-EQ'
> plugin, with a parametric section set to 1 kHz, -10 dB gain, and
> a bandwidth of 1.
>
> The 'Audio' track records the output of 'Bus 1'.
>
> Initial level is OK at -20 dBFS. At 9.5s the bandwidth of
> the filter is reduced to 0.1. This should not modify the
> output, but clearly something strange is happening.

The settling time for a filter with a bandwidth of 0.1 Hz is in the
order of 10s.  The existing filter state only represents about 1 second
of information.

> At 16s the bandwidth is set again to 1. The actual level of the
> resulting burst is around +20 dBFS, that is 40 dB more than what it
> should be. In other words, instead of say 0.1 W, your amplifier will
> produce 1 kW, assuming it has that power. Mine will clip at 400 W,
> still more than enough to fry the speakers.
>
> If this is representative of the quality of the a-*** plugins (by 'the
> Ardour Team') then I wonder what to expect next.

The mathematical problem as such is ill-posed when changing filter
parameters but I do agree that it would make sense to try transforming
the filter state in a manner compatible with letting audio hardware
survive and, if compatible with that goal, without changing the
steady-state response of center-frequency sinoids.

I'm not an Ardour developer myself but pretty solid at signal processing
(in spite of actually having an engineering degree).  I suspect that
you'd be qualified enough to fix this yourself but I could pitch in here
in return for some care on Aeolus (like applying patches and answering
mail).  Is the a-EQ the only filter you would want to see sanitized?

Feel free to answer in private.

-- 
David Kastrup