Iirc Gedlee was mostly interested in transducer distortion but it raises an interesting question of how to test that psychoacoustically, since it's hard to control transducer distortion the way you would want to in an experimental setting. You can't just turn up 'distortion' it's a complex byproduct of mechanical, thermal and acoustic reactions.
The idea that distortion is only a factor in transducers is an interesting one; I think it's a totally reasonable assumption that if distortion is at, you know, -120db or whatever, it's effectively 0, but part of me thinks that if distortion perception is mysterious and nonlinear enough, there may be some distortion which is objectionable even at trace amounts. Realistically, however, -120db is incredibly quiet.
A presentation by Earl can be seen here:
It has a good illustration as to why high order harmonics are more objectionable - the illustration implies it is because they are less likely to be masked since they are further away from the masking (fundamental) signal. It also illustrates the effect of level.
Some other fun takeaways:
- Audio compression can have very high THD but be considered audibly transparent. Audio compression encoders model masking effects.
- The test conditions were a .wav file of music, distorted using MathCAD
- Room effects were neutralized by the use of etymotic earphones
The correlation of THD and IMD to perception were .423 and .345 respectively; The Gm distortion model correlated far better (.94) It's pretty convincing.
Geddes conclusion is that since distortion in speakers is low at low levels and high at high levels, and since high levels mask distortion, it is likely always masked.
The most interesting results come from his tests of small signal delays, for example cabinet diffraction. Audibility of these effects increases as they impact linear distortion (not surprising, this is basically baffle step+ripple) but audibility of this distortion
increases with signal level. In other words, it is far more nefarious than THD/IMD from driver non linearity, which tends to be masked.
He advocates, as a result, for low diffraction designs, especially for very loud speakers. I can vouch for the transparency of my Gedlee speakers although I've never played them even close to their limits.