Maximum-likelihood catalog fitting yields accurate loudness estimates from few observations

Evidence

In this paper, we show how a catalog of MCPFs can be used to find the ML estimate of a listener’s MCPF, when a relatively small number of experimental trials are available. The ISO adaptive recommendation for CLS testing results in a relatively low number of experimental trials (≈ 15–30), in order to reduce the testing time and make the test more clinically realizible. Although this is an efficient approach, due to the low number of samples, the resulting CLS function can be inaccurate. The ML estimate of a listener’s loudness perception based on this lower number of experimental trials is able to more accurately predict a listener’s underlying CLS function, without removing outliers or using assumptions about the shape of the function to smooth the result. — Dijk, Pim van, Physiology, Psychoacoustics and…, 2567

The catalog can be used to compute a ML estimate of a listener’s MCPF. This may be particularly useful when the number of experimental observations is relatively low, but an accurate estimate of a listener’s loudness perception is needed, as is the case in most adaptive-level CLS methods that would be used in the clinic. — Dijk, Pim van, Physiology, Psychoacoustics and…, 2518

Categorization

• QUESTION: How reliable is ML estimation from the catalog when a listener’s true MCPF deviates from all catalog patterns? No validation metric is given here.
• corroborated: 4 independent passages support this claim — Dijk, Pim van, Physiology, Psychoacoustics and…, 2567; Dijk, Pim van, Physiology, Psychoacoustics and…, 2518
• XREF: connects to: Adaptive loudness scaling needs only 30 trials versus 440