The decoder is float-based and the test needs to allow for some fuzz.
Signed-off-by: Diego Biurrun <diego@biurrun.de>