Hi Lonce,
Thank you for your mail. I had a look at your demos and they sound really
good. The 'wind' is a good example of what we expect to be able to produce
soon.
Coming back to you mail, a shortcut would be "be modest" and I agree that it
is worth being reminded.
This given, MPEG4 is a great opportunity for us to design exciting
applications, encompassing natural and synthetic audio with environmental
spatialization, video, graphics, face and body animation, Text-to-Speech,
etc, etc.
MPEG4-SA is (and is recognized as such) one of the major achievement of
MPEG4 and it deserves that people pay attention to it when designing their
application. (BTW if you know any other work on MPEG4-SA in MPEG4
applications I would be interested to know).
At the MPEG4/BIFS level there is a node called AudioBuffer which is
specially designed to provide an interface to short snippets of sounds to be
interactively triggered. This is the functionality that we will be focussing
on first to include synthetic auditory icons in a scene. (That's why we are
looking for short sounds with high semantic value.)
We think that we can use that node also to play ambient sounds (wind, crowd,
sea) using the "loop" function.
Concerning my list of "points of investigation", I should have written
"potential points of investigation".
And last but not least, no we are not discouraged ...but any help or
comments are welcome.
Best,
Jean-Bernard
-----Message d'origine-----
De: Lonce Wyse [mailto:lwyse@mindmaker.com]
Date: lundi 31 janvier 2000 06:50
À: RAULT Jean-Bernard CNET/DIH/REN;
guillaume.fayemendy@cnet.francetelecom.fr; saol-dev@media.mit.edu
Objet: Re: sound FX generator
Hi -
At Mindmaker we have been developing something we call
FlexEffex. You can see/hear a couple of our interactive sound
effects at: www.mindmaker.com/demo/flex (Windows systems only).
We don't use MP4, but the idea is similar - downloadable
algorithms controlling arbitrary synthesis.
If I were asked to share the key insight of our work (and
even if I wasn't asked), I would say that even if you have
1) all the low-level audio synthesis routines you could desire,
2) a machine fast enough to run anything you write in real time,
3) great graphical tools with lovely little boxes to interconnect
with wires to put together algorithms without the "tedium" of
writing code,
you would still not be able to quickly produce an arbitrary
desired sound model. We (the all-inclusive one) simply don't know
how in general.
You would be able to do lots and lots of fun and great sounding
things - you would even be able to generate just the model you
want for certain classes of interesting and realistic sounds
which would give you confidence and an inclination to make
fantastic claims about your system. Then somebody would ask you
for a specific model they have in mind. - It would still take you
a long time to craft a new model that would morph between leaves
rustling and paper crumpling. You would still need good ears,
good intuition, a lot of experience and a bit of luck.
It is not just that realistic sounds are hard to generate
synthetically (even for the relatively simple class of musical
instruments!), but that a *model* is something that puts a whole
bunch of sounds into the same synthetic parameter space. "Specs"
come with demands not just on what sounds must be captured, but
the nature of the transitions between them (the desired
parameters).
I don't want to sound discouraging - in fact, progress in this
area is fairly easy to make (because there is so much of it still
to be made), not to mention that it is a heck of a lot of fun.
Best,
- lonce
This archive was generated by hypermail 2b29 : Mon Jan 28 2002 - 12:03:51 EST