You wouldn't use a "voom transformation" ... voom doesn't perform much of a transformation at all as it simply provides something like a +0.5 smoothed logCPM estimate for the counts form its inputted DGEList (though, I will grant that this is a transformation! :-).

The magic of voom is the "sister" weights matrix that it provides, and for that to be useful, your downstream method would have to be one that can leverage these observational weights.

You likely want some type of "variance stabilizing transformation" of your count data, though. In the edgeR/limma world, this would involve calling `cpm` on your count matrix with a value somewhere between 2-5 for the "prior.count" argument (sorry, but I can't give you better guidance on the choice of "prior.count" ... picking "the right" value for that (if there can be one) seems like a bit of voodoo for the time being, but perhaps Gordon can chime in), cf:

Alternatively you could use the "varianceStabilizing" or "rlog" transformations from DESeq2, see the "Data transformations and visualization" section of the Differential analysis of count data vignette in the DESeq2 package.