That difference signal experiment will not work well. At this resolution level, D/A and A/D conversion are not deterministic. Difference signals will be inevitable.
Your earlier experiment was similarly fundamentally flawed, even more so because of room acoustics and microphone uncertainty.