Note that if you're OK with white-noise only, you could consider generating it on the fly, thus requiring very little memory compared to storing several seconds of audio samples, and as long as your generator is decent, you wouldn't have a problem with periodicity, at least not on a few seconds scale. White noise is basically random noise. Just don't use the standard rand() function for this, as most standard implementations are relatively poor in terms of "randomness". I've implemented one of Knuth's pseudo-random generators to generate white noise in past projects, and it turned out pretty good. ("The Art of Computer Programming, D. Knuth, Volume 2.) Even a small MCU would have enough CPU power to do this in real time @44.1kHz or so.