Hi Carole,
The relevant outputs are:
_vox_p_tstat: uncorrected p-values
_vox_corrp_tstat: corrected p-values
_tfce_p_tstat: uncorrected TFCE p-values
_tfce_corrp_tstat: corrected TFCE p-values
All these are produced internally by randomise as Mike said. You can use either of the corrp files for your inference. Both are valid and control the family-wise error rate (FWE). They are based on permutations are not stringent, i.e., they are not like Bonferroni. In fact, they are used as a (much) superior alternative to Bonferroni.
The error rates obtained with a procedure that controls FWE can't be compared with those with a procedure that controls FDR, because these are different things. In any case, if you'd like to run FDR, you take the _vox_p_tstat or _tfce_p_tstat (that is, the uncorrected p-values), and run the command "fdr". There's no need for the --conservative option, as even with TFCE, the dependence structure is "well behaved" (as long as the preprocessing followed what can considered good practice, without global normalisations and things alike).
All the best,
Anderson