[sdiy] Hardware convolution box?

[mars at pingdynasty.com]

The main argument against straightforward convolution is the lack of 
parameterisation, which is what I take Frédéric to mean by 'too generic 
approach'.

Yes, convolution can be used for great sounding reverbs, and a good IR 
accurately captures the sound of any space. But it does so from one 
particular place, in one room, in one state (empty, humid...), recorded 
with one mic. You might get a 'wet/dry' knob but that's basically it.
And yes it can be used to accurately imitate any (linear, time 
invariant) filter, but you don't get to control things like cutoff 
frequency, resonance or anything else.
Same with cab emulations: one configuration, one mic position... It's 
like finally getting your dream amp, only to find that all the knobs are 
stuck!

There are of course tricks you can do to give the user more control, and 
cross-fading across a table of IRs would be one of them. But you are 
then no longer in the realm of straightforward convolution.

There are plenty of software convolvers, but most people (from what I 
can tell) tend not to use them for the simple reason that 'IR playback' 
just doesn't give you enough control over the sound.
Actually why would you want a hardware convolution box, rather than a 
laptop with unlimited IR storage?

Convolution still has a place IMO as one part of a more complete 
offering. The potential is massive, and I think there's an awful lot 
that can be done once the performance hurdle has been overcome and the 
technology is more accessible. It's like the early days of audio 
recordings; I say bring on the (impulse response) wavetable synths!
But if all you want is a wax phonograph, maybe check this out:
http://www.logidy.com/


Martin

On 2017-02-13 21:13, cheater00 cheater00 wrote:
> Hi Frédéric,
> 
> On Mon, Feb 13, 2017 at 9:49 PM, Frédéric (Opensource)
> <marzacdev at gmail.com> wrote:
>> Hi cheater00,
>> 
>> thank you for your didactic answer.
>> 
>> I clearly understand the maths behind using an impulse response to 
>> model
>> a  transfer function. It's basically converting an approximation of an
>> unknown
>> function into a big huge FIR or a polynomial (z domain) of an extreme 
>> high
>> order (as you prefer!).
>> 
>> I think, for many musical applications, this is a (too?) generic 
>> approach
>> because
>> it does not benefit from the context (what you are actually 
>> modelling).
>> 
>> You take the example of modelling "filters".
>> For audio DSP engines,  I often use IIR (or biquad) blocks in 
>> conjunction
>> with
>> various non-linear (well their polynomial approximation) elements.
>> How big has to be the equivalent FIR to get a satisfying similar 
>> impulse
>> response?
>> And how do you deal with alteration of the filter parameters (still an
>> example)?
>> Do you make a impulse response table (with interpolation ;-)) ?
>> 
>> Same applies for complex delays ...
> 
> I believe you're over-thinking this. A FIR filter will have its own
> sound, and that's it. It will be a different sound to analog, VA,
> component modelling, or FIR. Being able to change the cutoff in real
> time is not a challenge. I also think that the definition of
> convolution you're operating on is too low level. You're missing the
> forest for the trees. Convolution basically means "do stuff in the
> frequency domain" and "apply a volume and phase change based on
> frequency", that's it. You don't need to think about polynomials,
> FFTs, all that junk. That's too much conceptual load.
> 
>>>> Furthermore the platform could obviously support any sort of synth, 
>>>> be
>>>> it physical modelling, string, component modelling, granular 
>>>> synthesis,
>>>> FM, PM, VA, sampler
>> 
>> I understand the platform as being the hardware, am I right?
>> 
>> So what you ask is a generic box with MIDI ports, USB, audio in and 
>> out, a
>> very
>> beefy CPU or/and DSP, fast memory and with a well documented and 
>> accessible
>> SDK for anyone to develop for?
>> 
>> Is that alright?
> 
> Not really, no, we could just start by using dev boards, and see where
> that leads us, and then we can see if we can come up with a board that
> has some nice common features everyone would like to use. Some of the
> DSP dev boards have really good bang for the buck and unless someone
> proves to me otherwise I assume you'd be stupid to try and spin your
> own boards.
> 
> _______________________________________________
> Synth-diy mailing list
> Synth-diy at synth-diy.org
> http://synth-diy.org/mailman/listinfo/synth-diy