[LAU] Linux Audio podcast. episode003: commenting replies

[Fons Adriaensen]

On Sat, Aug 17, 2013 at 07:29:25PM -0700, J. Liles wrote:

> I completely agree. But I really think this is a more general problem. Most
> plugins are crap. That's a fact. LADSPA, LV2, VST, AU, whatever. Most of
> them are ununique, incomplete, poorly thought out, devoid of QA, etc. I
> think it would be generous to say that 10% of plugins are useful. But since
> when are we talking about plugins?

The topic of the thread was about reporting bugs. What should a
user do when he/she encounters things like described above ?

* To report a bug would be completely useless in such a case. *

...

> There's a whole different problem of branding/marketing and the
> misconception that there are even enough unique DSP tasks that anyone would
> require 100s of plugins. The truth is, anyone only needs a handful of basic
> plugins: the rest is permutations.

For normal audio processing (EQ, dynamics, effects,...) that is
true. If you count 'instrument' plugins as well things could look
different. But why should those be plugins in the first place.

...
 
> But I don't think this discussion was ever really about hardcore DSP
> programming. 90% of applications is user interface. And anybody can learn
> anything. Everybody starts somehwere. As charming as it may be to think of
> you this way, Fons, I doubt that even you were born already being a DSP
> guru.

I surely wasn't. And it takes a long time and a lot of effort
to build up both the theoretical knowledge and the skills. And
even more to understand the practical consequences of some of
the standard theory in an audio processing context. It's also
a process that never ends. Just yesterday I discovered some small
but systematic errors in the white noise generator used in e.g.
Jaaa. [*] 

Anyone who's learning this sort of thing and who uses his/her
knowledge to write free software will have my full support.
But I won't do anything to make life easier for those who
just copy some DSP code and wrap a 'shiny GUI' around it, with
as the only result one more that just looks as much as a toy
as most of them do.


Ciao,


[*] Set up Jaaa to measure its own noise output, with averaging.
Zoom in on the vertical axis. The result will never be exactly
flat (as it is a finite lenght average), but the remaining 
errors should be random. If you wait long enough and repeat
this a few times, you'll notice that the small dips of 0.3 dB
or so at around 1, 2 3 kHz are systematic. This was a surprise,
as the noise generator is based on a random number generator
which has a very good reputation - whatever autocorrelation
it hase will probably be insignificant in most applications.
Yet this test will reveal them.

Will be fixed in the next release of course.


-- 
FA

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)