Percentage distortion metrics include inaudible masked components and mispredict perceived quality

Evidence

Thus began the legend that harmonic distortions are relatively benign and intermodulation distortions are bad. It is true in the context of these test signals, but they are both simply different ways of quantifying the same problem (the non-linearity in the audio component), and neither test signal (one tone or two) is even a crude approximation of human voices or music. Contributing to the mismatch between perception and measurement is the fact that such a technical measurement totally ignores masking. Included in the numbers generated by the measurements are distortion components that are partially or completely masked. Some of what is measured is inaudible. The percentage distortion numbers cannot accurately anticipate the audibility of the distortion products. — Toole, Floyd E, Sound reproduction- the acoustics and…, 990

Categorization

• XREF: Directly parallels the critiques of objective audio metrics elsewhere in this book and in perceptual audio coding literature — measured numbers vs. perceived quality.
• ADDENDUM: Relevant to the theme of measurable-but-inaudible degradation; connects to how measurement without masking knowledge misleads in audio engineering.