I'm making all my music on a modular, but often bring in "noisy" samples that I generated or processed on my linux laptop. Obviously offline rendering of "noisy" samples is not the same as a dynamically controlled plugin in ardour or the like, and what I look for is not as much a static noise spectrum/waveform as musically useful but noisy samples. With that in mind here's a few strategies I had success with:
1) Databending non musical files, by forcing sox to interpret them as audio. Works great with binary files (/usr/bin is a goldmine) and requires quite a bit of work afterwards to select the interesting portions of the samples. The middle section for instance tends to be more or less white noise.
2) Cecilia is always awesome.
Hope that is helpful.
Atte