posted 18 August 2001 13:40            

NewGranulatorTechnique.kym

First, some caveats. I have a Capy 320 with two cards (8 DSPs), so I am not certain if these will work on the basic system. I am still getting odd scheduling results with Kyma 5.16, so although the DSP meter looks like they should work on 4 DSPs, I wouldn't bet on it. You may need to use ForcedProcessorAssignments in some cases.

There are three versions of this Sound - one for samples, live input and disk files. The granulation method is identical in all three. I only create eight grains, but this seems to be enough in most cases, and the Sound was getting too complicated anyway.

Each grain is created by multiplying a Sample with a FunctionGenerator which has the grain wavetable (Bristow-Johnson). The grains are spaced evenly in time, there is no jitter. The length of the grain changes with the rate, it gets longer as the rate slows, and it can also be varied manually with the grain shape control. An oscilloscope display has been added to allow viewing the shape and size of one grain. The grains can be up to 8 seconds long, which I found to be very useful.

The start location within the Sample can be varied two ways. There is a manual control, Start, useful for "freezing" a sound, and there is a "Scan Mode" which automatically moves the start index through the sample. The rate of the scan can be varied, and the direction can be reversed. The "Scan Location" rotary is an indicator (NOT a control!) showing where the Scan Mode oscillator is in its cycle. The Reset button resets the scan oscillator to zero.

Both of these controls for start location are global, they effect all grains the same way. There is also a Start-Jitter which randomizes the start index for each individual grain. The Start-Jitter applies to both manual and scan modes.

The grain Pitch can be varied by a global multiplier, which shifts all eight grains the same amount, and by Pitch-Jitter, which randomly adds or substracts from the pitch on an individual grain basis. There are also Pan and Pan-Jitter which work similarly.

Finally, there are a pair of compressors on the output and an oscilloscope to view the output.

All of the above applies to all three of the Sounds. Now for some differences -

The Sample version - use this one to play samples which are shorter than 180 seconds. This WILL work with stereo samples, but I have to use ForcedProcessorAssignments to make it schedule properly. If you use stereo samples and set the Pan to 0.5, you will get stereo results. Given what you are going to do to it, mono samples are probably fine anyway.

The Disk version - use this one for files greater than 180 seconds. It is NOT stereo. It will sum L+R and write them into a 30 second buffer with a MemoryWriter. The Sample reads that buffer.

The Live version - similar to above, the sum L+R is sent to a MemoryWriter. You can control the length of this buffer by editing ?dur in the Script.

Be aware that the MemoryWriter will continue to write to the end of the buffer even if you un-check the Record button. Also, there will be junk in the buffer if you stop and then re-start the Sound.