Kyma FFT vs MSP pfft~ ?
I was wondering if anyone knows how the msp fft compares to kyma?
does anyone know what the limitation of msp are in regards to this - it sounds good to me but there seems to be a lot of hype about kyma and dedicated hardware...
what's missing in the msp implementation?
one main difference is that in maxmsp you can not use more than 100% computing power in nonrealtime mode.
but one rarely needs more than 2000 bands for most applications imo.
right - thanks for the reply Roman. do you know how it compares in realtime mode?
i thought they'd actually be same in non-realtime as the desktop cpu can just take as long as it needs to compute.
people have mentioned patented algorithms for realtime before but i figured as convolution just involves multiplying the fft data (afaik) that the difference would have to be in the fft implementation itself and the kind/amount of data output..
maybe a larger window size without the time 'smearing' effects perhaps, but either way its still got to be the same fft mathematics
whether its a custom max external or dedicated hardware right...?
i guess it depends what you are doing and what computer you have.
with the new hardware, which i dont have, on kymaX you have probably
lots more room.
well and as for simple convolution, CPU is not a big problem anyway i guess.
i can do 15 channels of that in maxmsp on my 10 year old mac. :)
interesting differences probably begin when you start doing realtime control
over single bins for your custom special effects.
custom max externals is a consideration if you think you dont like pfft.
Another major difference is that Kyma comes with a nice integrated UI
timeline that allows you to sequence objects and automate changes.
The closest thing we have to that is Max4Live+Ableton Live, which
in the end still requires you to convert your patches to M4L instruments.
It is another layer of work that gets in the way of creativity.
Don't get me wrong, I love M4L it is a great enhancement. But in my
opinion Max still needs an integrated time line that lets me sequence
symbols, data, lists, etc.
ok - thanks. pfft sounds good to me but i haven't heard what kyma can do in terms of convolution based cross synthesis.
the topic of custom externals for fft brings up the same question of what else is out there that could bring about a better or different sound?
particularly in regards to realtime convolution / cross synthesis...
i´d say fast fourier is not exactly why people buy a capy/paca, it is more or less the same in all worlds.
compared to some of the resynthesis based kyma sounds cross convolution is uninteresting.
even the setup of a good biquad or FIR based vocoder effect is better (and simpler) in kyma when
compared to max. but fft is, well, fft.
also as well, the kyma systems do run off of dsp units. using outboard processors to help in spreading out the load from computer and hardware. much like pro tools actually. using hardware to do most of the leg work and the computer to run the software. so there is a big jump from using just software, to using integrated hardware as well.
i think kyma would be greater in terms of that, but really depends on your personal preference. i would love to have a kyma system myself. not just to have something better or whatever, but more to try something new, see what i can come up with. its just the price that does stop me :)
aha - ok, thanks for clearing that up. if anyone else is in the same position as me and looking for different techniques to work with fft data in msp beyond the tutorials, i recommend finding the papers out there by Zack Settel & Cort Lippe.
they've got a set of interesting operations you can do on the fft data.
i guess its not as good as kyma but i got the sounds i needed after implementing some of these techniques.
Hi srs,
Very interesting. Can you send us a link where I can find Lipp's writings about fft?
Thanks you!
done a quickie search on google came up with this link
www.music.buffalo.edu/faculty/lippe/pdfs/fft.pdf
for an excellent timeline that lets you sequence messages in Max, try timeline.js by Alexander Schubert (with Georg Hajdu):
or here: