more on Kyma.....

[Jim Coker <jcoker@interaccess.com>]

As Carlos mentioned, the Kyma software can build sound
algorithms that use any number of synthesis techniques.
One of the really nice things is that their software is
designed to make building algorithms fast and easy.  Most
operations are based on a drag-and-drop metaphor, and 
there are many useful shortcuts.  Another thing about their
software that is nice is that, unlike MAX, in which almost
everything has to be set up graphically (which can be *very*
tedious), Kyma allows the programmer to express some parameters 
in mathematical expressions, or even write Smalltalk scrips to 
modify or generate sounds.

In reference to Matthias' questions, most parameters on sound
objects (such as delay times, frequency settings, filter bandwidth
and frequency) are controllable in real time, and without glitches.
Put an lfo on a short delay, and you get a flanger. (Delay lengths can
be specified in a number of ways, such as in seconds, or samples,
or relative to something else)  Control signals can come from midi,
or all sorts of other things, frequency trackers, envelope followers,
other audio signals, or signals built by processing other signals.
When I was there, a Peavey PC1600 fader box (16 programmable
midi faders and assorted buttons) was set up to
control sound parameters. This is what I used during the looping
test (no foot controller available).

One interesting patch used an evelope follower to control a 
pitch shifter, the louder the signal, the higher the shift.  
The result, when combined w/ a few filters, was an excellent
"cartoonization" sound that with one filter setting
made the speaker sound like goofy, and another setting made it
more like daffy duck.

Another nice patch involved a "harmonic resonator", a special
kind of filter that resonates at a given pitch and all it's
harmonics.  It worked great w/ guitar, as you could 'play into'
it, bending notes so that they resonated as they were bent
into the resonated pitch.

Probably the most unique capabilities of the system revolve around
it's analysis and resynthesis capabilities.  Their latest software
version comes w/ a configureable vocoder w/ up to 70 filter bands.
In addition to real time vocoding, you can analyize a sample 
(drums, vocals and animal sounds work best, due to their widely
varying formants) to build a time-varying filter bank, and then use
this filter bank to process a live signal.  If the analysis sample
is speech, and the filter is swept at the 'original' speed, the
result sounds like a regular vocoding, but if you control the 
time sweep of the filter bank with, say a midi CC pedal, then you
have a very customized sort of wah-wah pedal that you can rock
to move forwards and backwards through the filters at will.
(How about a meow-meow pedal? or a barking pedal, or a 
"Dammit" pedal, or a ...)

For really hard-core stuff, you can use an FFT analysis to convert
the signal from the time domain into the freqency domain, and do 
processing there (such as stretching or scaling harmonics, pitch
and time shifting, etc), and then resynthesize the result using
an oscillator bank.  This is the approach used by Digital Performer
1.7 and others to do pitch shifting w/o ugly artifacts.  Kyma
can do this in real time, minus a 1/4sec delay due to FFT
windowing issues  (For a clear explanation of this, look for the
Curtis Roads book I mentioned in the earlier post).  Unfortunately
it takes a Capybara w/ 5 to 6 cards to do this, but you can do
the analysis on a sample ahead of time, and do the processing
and resynthesis in real time and use fewer resources.  The big
FFT takes about 2.5 cards (according to Kurt Hebel) and you can
get about 51 oscillators on each card for resynthesis.

Another sound of interest is the waveshaper, which can be used
to introduce new harmonics into a signal. Waveshapers are great
for producing distortion, and the nice thing about the Kyma one
is that the waveshaping function can be controlled in real time.
In one patch, they had 8 sliders set to control the coefficients
of the function.  A bit too much for playing, but pretty good
when searching for sounds.

The only dissapointment I had was with the frequency tracker.
It works amazingly well w/ vocals, but didn't do so great on
a guitar.  The response time was at least as good as a Roland
GI-10 midi converter and it did track vibrato and
other pitch nuances very well, but it would often guess the wrong
octave, and get confused by string noise. However, this was only a
first try and given some tweaking w/ filters and such, I bet
that it can be improved to the point where it is accurate 
enough to be really 'playable'.  One big change that would make
it better would be to use hexaphonic input a-la GK2, which would
restrict the pitch guessing range, and avoid multi-string noises.
Currently Kyma only has 2 inputs and 2 outputs, but they are working
on increasing this.  They get many requests to increase the 
number of outputs, but Kurt said that this was the first time they
had a solid reason for having multiple inputs, and that they
would prefer to do an expansion w/ a balanced number of inputs
and outputs.  Carla, who plays harp, thought about it a bit and
said that she really would like to have a 'digital' harp...

Also as Carlos mentioned, it is expensive, $4400 for a 2 card 
system (the price slowly is going down, very slowly), and $600
for each expansion card.  Each card has a 66Mhz 56002 w/ 3MB
of RAM (all processing is 24 bit), and each card's ram can be
expanded ($100 to push it to 12MB).  In terms of the processing
power available, I was able to get a simple (2 osciallators,
filter, 2 envelopes, mixing and lfo) 4 voice synth running
on a card, and I think an 11-band vocoder will fit on a card.
It is not as efficent as a dedicated device, but is much more
flexible, and does compare well w/ other music toys in its
price range.  It doesn't have all the neat programs that would
come w/ a top-end Eventide box, but is a lot more flexible.

Kurt and Carla both very much believe that an external DSP
mainframe is preferable to processing using a general computer,
and I tend to agree.  Even though the DSP's are slower
than the chips in new PC's, they are more efficent at processing
audio and don't have to bother with the operating system overhead.
Also, the capybara is expandable, and they plan to support faster
chip speeds w/o having to force users to get a new mainframe. You
could buy a 200Mhz PC or Mac and run CSound on it, but for serious
programs it would consume all the available processing power, and 
would not be very upgradeable.  Kyma can run simutaneously w/ a 
sequencer or MAX on a modest MAC or PC.  Symbolic Sound is also 
working on a PC-card interface so you can use Kyma w/ a laptop.  

Probably the closest competitor is MAX/FTS from IRCAM in France.
It is a version of MAX that can handle audio information.  The
audio processing bits are compiled and loaded onto custom
hardware for processing (just like Kyma), problem is, they keep
changing the hardware platform (first a Next box with intel
i860 chips, then SGI, then PC's, and the latest rumor is that 
they're back to using Macs) and it's not a real product, but
an ongoing research project.  Kyma has been evolving for 10
years, and their software shows it, both in quality and depth.

If you're really interested in more about Kyma, you can get the manual
for $35+shipping, and there's also a good review in the July '95
issue of Electronic Musician, and, of course, they have a web
site.

jim