As far as I know reads are never mapped to a GTF (or GFF) file, they are always mapped to a fasta file (or an indexed version of a fasta file), so that's a file with the actual sequence in it. The GTF/GFF do come into play when you want to link your mappings to annotated features (eg. genes, promoters, lncRNA, ... ) .
The best you can do is to map all your reads to the human genome.
As an alternative, you could subselect the human genome to only contain the regions you might be interested in. Based on you bed file you could do this with eg
getFasta from the
bedtools package . DO keep in mind that this has a huge potential in biasing your analysis.