Harmonic distortion measurements poorly predict audible annoyance from real music.

Evidence

The problems originate as measurable input/output non-linearity in devices, but the industry has chosen to evaluate the magnitude of the non-linearity by probing them with well-defined test signals and measuring the amount of added distortion products at various sound/signal levels. The simplest test signal is a pure tone, and the distortion products of pure tones are harmonically related overtones; therefore, we have “harmonic distortion”. Music is not simple pure tones, and the broad spectrum of music that generates the distortion products also makes them less audible—they are perceptually masked. As a 22. result, harmonic distortion, as a measure, is not a reliable indicator of its audibility. To design engineers, it is a useful relative measure when designing systems within their context, but as an indicator of potentially audible annoyances for consumers it is almost useless. The only truly safe value for harmonic distortion is zero, or very close to it. — Toole, Floyd E, Sound reproduction- the acoustics and…, 109

Categorization

• XREF: Directly connects to the perception of masking literature; single-tone tests are unrepresentative because broadband signals mask distortion products. — connects to: Judged pleasantness does not track the physical quality of a sound — Almo Farina, Soundscape Ecology: Principles,…, 1078
• QUESTION: What alternative test signals or metrics (e.g., intermodulation distortion, noise-based probes) better predict audibility?
• DEEPTHINKING: An industry-wide test metric (THD) that is ‘almost useless’ for its intended purpose, yet persists as a design benchmark—worth sitting with as a case study in how measurement culture outlives its rationale.