The concept of "read depth" isn't applicable to RNAseq, since it doesn't actually tell you much useful. This term is useful in calling variants, since you expect them to be more or less equally covered...so a single metric, like read depth, can convey useful information. For RNAseq, you expect there to be vast differences in coverage between genes/isoform/whatever, so there's no single number that could convey anything useful in that regard. If you must give a number for some reason, just give the number of mapped reads above some MAPQ threshold. Alternatively, do as Irsan and just use featureCounts or htseq-count and then attach the resulting count table as a supplemental file (afterall, this contains the actual useful information).

The cumulative non-redundant size of the human transcriptome varies from cell to cell and depends on how much evidence you find enough to consider something to be transcribed. Why don't you assemble the transcriptome of your own samples and calculate the size yourself? Or even better if you are just interested in read depth, use dedicated tools that calculate read depth from rna seq bam files directly.