:
:
:
:
:
:
:
:
:
:
:
:
:
:
::
:

:
2020

KYBER - GRANULAR CONVOLUTION PLATFORM

from public info and emails with a member of the design team this is what i can gather about how it works:

user records a sound snippet to one of 20 slots (max length 20s per slot)
the sampled audio from current slot is fed into a granular engine
select a grain (ranging from 7-700ms in size) which is then convolved with the real-time audio input
the spectral overlap of the two sounds is output (think…grain timbre w/ input envelope)
additional details from the devs:

the convolution is done using Convolution2L since it allows for seamless transition between grains (convolution kernels)

it doesn’t support proper time-partitioned convolution so to keep the input-output delay small they manually delayed the input signal and fed it to several Convolution2L that all ran at a small framesize (& got a different part of the grain as convolution kernel)

two challenges they faced [1] each grain needs to have an envelope applied [2] the energy of the output needs to be controlled…

they used a short single sided hand window to fade in/out

they ended up normalizing with the RMS of the grain (capped at some higher / lower limit)

Any time a new grain was selected (either by shifting the position of grain window or by changing its length), it was prepared (applied window, calculated gain, loaded into all convolves) and then a trigger signal was sent to all convolvers to switch to the new grain

___
CONTROLS

arc enc 1: size
arc enc 2: pos

other assignable params...
input vol
output vol
sample slot select

interface...
horiz dots for slot selected
vert dots for input vu meter
bar for grain

steal from...
mangl
sway
timber
nisp
cranes

https://fairfieldcircuitry.com/products/shallow-water