For example, they give us an audio and they tell us a robotic voice is telling "Laurel", but they then tell us that what we hear depends on the speakers.
So the fact is that we are not hearing the same thing. So it is not a valid test.
Indeed, the test is wrong because it is computer synthesizers the ones who have problems pronouncing certain sounds, like the "l" that we create putting our tongue back and creating a loop that lets air go out by the sides.
This is a complex analog movement and resonance that is hard to emulate with digital technology, because it is not static and dynamic sounds are very hard to control as we use Fourier analysis for that with windows that introduce distortions in the signal.
Most companies out there just record those sounds from real people.
So those guys instead of accepting the limitations of current technology and accepting their test is wrong, they use it to prove that others are wrong while hearing it.
Oh, the irony. The interesting thing is that while it is very hard to admit that you are wrong, it is extremely easy to admit that other people are wrong.