I'm not using version 2.0 because that version makes it almost impossible
to create a slim build without compiling all the voice synth code (which
we don't need).
Note that I'm using a hacked Google Test version.
It is based on commit 13206d6f53aaff844f2d3595a01ac83a29e383db from 2015/12/23,
with pull request #621 manually applied to fix the build.
I'm planning to switch to release 1.8 as soon as it's available.