I suspect trying to reduce the jitter in a USB keyboard response could be frustrating.
I have some experience conducting experiments that are structurally similar to this. To really be accurate, how about using a piano keyboard, or some other physical sound-producing device. For each test record a stereo track of the whole test. On one channel, use a microphone to record the stimulus that the subject is hearing. On the other channel, record the output of whatever device the subject has been instructed to use.