gen~ version of grooveduck

Here's that gen.slicer example updated to support timestretching of a buffer~.

Notice how the window duration affects the sound. On a transient rich, unpitched source like the drumLoop, it sounds pretty good with very short windows, 10-500 samples, imparting a kind of phasey artifact. On a pitched source however, such a window sizes conflict with the pitched material in a kind of amplitude modulation way; instead, larger window sizes in the thousands of samples are better, leading to less obtrusive echo effects.

The phasey/pitchiness of the artifacts partly comes down to the regularity -- the window size is always the same and repeats with a constant frequency. So the regularity could be reduced by modulating the window duration. This is demonstrated in the gen~.pitchshift.maxpat example, which adds randomized offsets to window sizes for each grain. (This example also shows how to layer four rather than two windowed samples to smear it even more). But it kind of just replaces the ringiness with noisiness.

What I was suggesting is that, if we can tell if the input is more transienty or more pitchy, pick a window size accordingly. For pitched material, an ideal window size would be the fundamental pitch, as this would avoid introducing new frequencies to the output. One of the simplest methods of detecting pitch is to measure the time between zero crossings. It's not a great pitch detector, but for fairly simple waveforms it's not bad considering how simple it is. So, the idea is, if we can guess what the period of the input waveform is, we can align windows to that same period, and the artifacts might not be very noticeable. For transient-rich waveforms this period is going to be really short, so it might just work automatically.

Graham