No, I don't think it is wise to use Bracken on metatranscriptomic samples. Bracken uses a Bayesian approach to re-estimate species abundance of reads belonging to closely (and largely similar at the sequence level) species. An assumption of Braken is that the number of reads identified to a particular species is proportional to the abundance of this species on the sample, which is reasonable to metagenomic samples. However, this is not a reasonable assumption for metatranscriptomic samples, as genes have varying levels of expression on a cell, and these levels are not necessarily consistent between species.

Anyway, Braken don't recover any reads, it just estimates species abundance based on read classification by Kraken.