I have not tried bit-depths other than 8 and 16, though I don't see much need to. There isn't really a point to using a non multiple of 8 bit depth, since the integer values have to line up on byte boundaries (i.e. if you had a bit depth of 13, you'd be wasting 3 bits per sample frame, or 6 if you have 2 channels). Bit depths of 24 and 32 are not unheard-of, but imo the slight increase in quality is pretty much imperceptible and not worth the extra bytes.