Psychoacoustic loudness models cannot accurately predict individual listener responses

Evidence

Substantial research over many years resulted in much more elaborate methods for computing loudness as described in Fastl and Zwicker (2007) and Moore et al. (1997). They begin with narrow-band spectral analysis and loudness-summing procedures based on psychoacoustic models of some the frequency, amplitude and masking properties of the hearing process. The complexity of such models appeals to some, but the supporting data are limited and include large statistical variations in subjective responses. As stated by Moore et al. (1997) in section 3, describing limitations of their model, “the predictions of the model should not be expected to be accurate for individual listeners or even small groups of listeners”. — Toole, Floyd E, Sound reproduction- the acoustics and…, 1175

Categorization

• XREF: Relates to the reader’s broader interest in psychoacoustic modeling limits; Moore’s own caveat connects to the general problem that model accuracy falls at the individual level, echoing themes in other psychoacoustics texts.