http://www.nabble.com/Modal-Freqs-tf4536626.html

Hi folks,

I'm picking up the recent thread from the Blue list & posting on the Csound
general list as i'm not actually subscribed to the former.

Firstly, a simple question: How is mode any different from any other
"generic" filter then? - it's essentially just self oscillating under
extreme Q settings, yes? or is there more to it than this? could i
experiment as effectively / in a similar manner with other filter types?
(effectively, by extension, things like Karplus strong are just a form of
"filter" anyway - am i correct in saying this?)

Secondly (& somewhat longer & subjectively) I know i often seem to be
waffling on about several types of resynthesis all at once in my posts, but
having continued to think about this through the night (it seems if my
recollections of sleep are accurate) essentially i think i have kind of been
able to summarise why this is.

It seems that the data retrieved from some types of analysis is better
equipped to allow "extensions" to be calculated / interpolated from it than
others (ie model 3 octaves of Koto from 1 or 2 koto samples by randomising &
rescaling the data components) 

for e.g. Loris is probably "good" in this dept (particularly, say, morphing
between a low octaves sample & a higher octave note to produce the
intervening range - I haven't been able to try it) & PVOC, probably "not so
good" for this kind of "realistic modelling by extension" (but very good for
lots of other things obviously...)

So what i'm groping towards is "why not actually use several techniques
together" somehow (i.e a bank of sines, Band Limited noise, AND resonators
for example) to more effectively achieve both 

* a more convincing resynthesis, & 
* a greater capacity to do what i am describing above (deduce wider
instrumental ranges from limited numbers of source samples...)

Particularly when (in the case of the Mark Van P's Xylophone example i link
to above...) I'd hazard a guess that his "modal freq's" are pretty close to
any prominent partials i might find in a SPEAR / SDIF type partial analysis,
ditto prominent frequency bins in a PVOC analysis.....

So the point is then surely, equipped with data about "prominent freq ranges
& their widths" - the question as to how they are reproduced could work in a
much more flexible way, & i could possibly resynthesise ,a Koto say (or
Xylophone or whatever), more effectively by using a couple of resonators, a
couple of sines, & some band limited noise to various degrees depending on
the application? (ATS after all uses 2 of the 3, but doesn't allow for noise
"transposition" - does it allow noise "morphing? - anyway, at present i
can't resynth it anyway....(whinge whinge...))

Resynthesis data doesn't have to be "directly translateble" from one process
to another, but surely the correlation between resynthesis data types is
maybe stronger than is mostly implied by the "analyze using technique X,
resynthesise using the same technique" approach? But it seems many
resynthesis techniques don't allow us mere mortals to get under the hood &
deduce this type of information. Is it not possible to approach resynthesis
as more of a "statistical exercise" rather than a "pop the coin in the slot
& pull the handle" one?

I know, I'm not really "asking" anything - but does ANYONE understand what
I'm saying & what i'm trying to work towards here? I think i'd go as far as
saying a SPEAR type SDIF analysis is the most "transparent" of analysis
types & display in terms of providing data that is most open to "free form /
combination" resynthesis ..... perhaps now i just need to get a greater
understanding of mode & how it works & experiment with the techniques &
processes I am describing...?

None the less, i welcome thoughts, experience, encouragement, & believe &
hope there is something here worthy of consultation (or at least reflection)
for those far more intelligent & well versed in these techniques & their
underlying basis & inner workings than myself... & hopefully further
progress (by all of us) can be made towards accurately modelling acoustic (&
"acoustic like") sounds...

regards

T.


-- 
View this message in context: http://www.nabble.com/modal-freq%27s-%28cont%27d%29---more-resynthesis-pinings---wishes-tf4616852.html#a13185390
Sent from the Csound - General mailing list archive at Nabble.com.

Quote Ugur:

# IMHO the difference is in our mental model of the system. A filter is
something which modifies a signal, which is an abstract mathematical
operator, but a mass-spring-damper system is something which reacts to an
input force, which is a physical thing.#

Absolutely. Ugur, thank you for such an excellent & thorough reply - that
with a bit of effort on my part it will also serve to gently break my
ignorance over some of the maths behind filter designs...

I guess what i'm overcoming here (& in general) is the divide between these
"physical systems" i want to try & simulate, & the dsp systems required to
do so. At some point, once you are "under the hood" / messing with the inner
workings, it's definately more meaningful (for me) to "switch" my thinking
on the issue. By doing so - it exposes alternate options & processes for
exploring the same "modelling" outcomes.

Your post also draws the concepts of resonating filters & convolution closer
together in my mind, which makes sense given the relevance of the latter
also to modelling instrument bodies.

I had never thought of a FFT based spectral delay unit for example as an
alternate & more intuitive way to design an allpass filter before, but
assumedly that also must probably the case....(& in fact pretty much what
"spectral delay" is?)

with so many possibilities & options on offer, i think there comes a point
when this kind of "homogenisation" or reduction of thinking about processes
becomes valuable. It makes the totality of available options easier to
overview, & simultaneously opens up more avenues of exploration through
identifying similarities between processes that, superficially, seemed quite
different.

As for dealing with the same set of issues regarding the different
resynthesis processes, well, wasn't necessarily looking for an answer to
that one - more just trying to make developers & people aware about the kind
of "advances" i'd personally like to see being made (or at least
contemplated) in that area....

Many thanks again Ugur

T.

-- 
View this message in context: http://www.nabble.com/modal-freq%27s-%28cont%27d%29---more-resynthesis-pinings---wishes-tf4616852.html#a13205520
Sent from the Csound - General mailing list archive at Nabble.com.

Tim Mortimer wrote:
> http://www.nabble.com/Modal-Freqs-tf4536626.html
>
> Hi folks,
>
> I'm picking up the recent thread from the Blue list & posting on the Csound
> general list as i'm not actually subscribed to the former.
>
> Firstly, a simple question: How is mode any different from any other
> "generic" filter then? - it's essentially just self oscillating under
> extreme Q settings, yes? or is there more to it than this? could i
> experiment as effectively / in a similar manner with other filter types?
> (effectively, by extension, things like Karplus strong are just a form of
> "filter" anyway - am i correct in saying this?)
>   

My guess is that historically instrument builders made instruments that 
had a nice sound, which in certain cases turned out to be something that 
can be modeled with the mass-damper system, that can be imitated with 
the mode opcode. Other filters probably also imitate sounds that are 
physically possible, but less interesting. Hence the interest in the 
mode filter.

> Particularly when (in the case of the Mark Van P's Xylophone example i link
> to above...) I'd hazard a guess that his "modal freq's" are pretty close to
> any prominent partials i might find in a SPEAR / SDIF type partial analysis,
> ditto prominent frequency bins in a PVOC analysis.....
>   

You refer to one of the worst Csound instruments I have created :-)

I have no experience yet with the SPEAR / SDIF type partial analysis, so 
I don't know if it can easily find the modal frequencies. The mode 
opcode amplifies a range of frequencies, and the input signal (such as 
produced by the pulse opcode) typically contains all frequencies, so a 
continuous range of frequencies is present. So an analysis that looks 
for a limited number of frequencies may fail here. But I could be wrong 
here.

-- 
  Mark
  _________________________________________
  When you get lemons, you make lemonade.
  When you get hardware, you make software.

Hi Mark

I guess with the Xylophone example 2 things i considered (briefly)

1) when i analysed something similar using SPEAR (Marimba i think) there
were a few very prominent partials, & that was pretty much it.

2) as the PRAAT analysis still produces a freq value as a "centre" of the
resonant range - assumedly there would also be (in a SDIF/ additive style
analysis) a similar partial viewable at this frequency. Similary any
resonance width would possibly show up as prominent partials either side of
this centre.

In essence, that's why i used it as an example (& the fact that it started
me thinking in the first place, along similar lines to what tends to run
through my head fairly regularly)...

As long as just about any resynth data is viewable & analyseable /
summatable / "reduceable" to some degree, & then viewable & editable -
whether a sound was analysed using PRAAT / SPEAR / whatever - the data
obtainable could then be used to resynthesise a sound by pretty much ANY
alternate means - as long as it could be RE presented in the alternate
format somehow - if not exactly, then at least in an analogous sort of way.

i.e if i see a wide band of partials in a SPEAR analysis - i could then say
"great - treat that as a "centre of resonance", & model that part of the
spectrum with a Mode, or a Loris code", or whatever...

similarly, if PRAAT yielded "narrow bands of resonance", then perhaps an
additive resynthesis would suffice just as well....

By extension, this process could apply across a whole range of synthesis &
resynthesis methods, potentially to be then used in combination to more
realistically modeal acoustic sounds & instruments.

having analysis data "universally accessible" enough to achieve this would
also mean that data could be randomised, or parametrically re-calculated to
"imply" things like velocity sensitivity, or wider instrumental ranges than
any original analysis of individual samples might itself offer.

& this is ultimately where i would like to head using things like Csound. As
a "wannabe post - everything music god ; ) " i can only look at the levels
of computer based simulation in films & the visual world, & can't help but
conclude that in some respects the world of "acoustic simulation" is lagging
maybe 20 odd years behind?

"I have a dream..." etc etc.....

If you want to upload that Xylo sample somewhere i'll do a SPEAR analysis, &
we can compare the data perhaps, as a little experiment....

in a bit of a rush at the mo. hope that has some worth / clarifies further
my impetus here...


Mark Van Peteghem-2 wrote:
> 
> 
> You refer to one of the worst Csound instruments I have created :-)
> 
> I have no experience yet with the SPEAR / SDIF type partial analysis, so 
> I don't know if it can easily find the modal frequencies. The mode 
> opcode amplifies a range of frequencies, and the input signal (such as 
> produced by the pulse opcode) typically contains all frequencies, so a 
> continuous range of frequencies is present. So an analysis that looks 
> for a limited number of frequencies may fail here. But I could be wrong 
> here.
> 
> -- 
>   Mark
>   _________________________________________
>   When you get lemons, you make lemonade.
>   When you get hardware, you make software.
> 
> -- 
> Send bugs reports to this list.
> To unsubscribe, send email to csound-unsubscribe@lists.bath.ac.uk
> 
> 

-- 
View this message in context: http://www.nabble.com/modal-freq%27s-%28cont%27d%29---more-resynthesis-pinings---wishes-tf4616852.html#a13227257
Sent from the Csound - General mailing list archive at Nabble.com.