Why randomize the phase? about paulstretch_python

[paulnasca]

Hi.
One reason I randomize the phase is that the resulted sound is more soft (more pleasant to listen many hours continuously) and also the algorithm is much simpler even if I sacrifice the transients. Even more, usualy I recomand using Paulstretch with larger windows (more than half a second) resulting subjectively very pleasant sound texture.
Many years ago I tried myself, too, using phase differences and other ways to interpolate the phases, but I didn't liked the results.

Another reason is that phase randomization models very good for ensemble sounds (like choirs, orchestras). I described more about it on ZynAddSubFX page: http://zynaddsubfx.sourceforge.net/doc_0.html

Anyway, the "paulstretch_newmethod.py" it have an onset detection which changes the speed on onsets but I sill like more how the "paulstretch_stereo.py" sounds (the original algorithm).

For the video you gave me, indeed, it is very accurate stretch (more accurate than Paulstretch could do in relation to the transients), but do you have an choir processed with the same algorithm? I am curious how it sounds on this ( http://hypermammut.sourceforge.net/paulstretch/original.ogg ), for example, at 20x stretch.
Thanks.

Paul