Optimised adaptive procedures and analysis methods for conducting speech-in-noise tests

Figures & data

Table 1. Adaptive stimulus placement procedures investigated in this study. The prior and parameter and stimulus spaces are specified for each respective method.

Procedure	Prior	Parameter space	Stimulus space
Reference	N/A	N/A	[−20,20]
modReference	N/A	N/A	[−20,20]
Interleaved	N/A	N/A	[−20,20]
FIG	Flat	α = linspace^a(−10,20,81)	linspace(−20,20,81)
		Width^b = linspace(2,30,11)
		γ = 0.02
		λ = 0.02
Greedy	Flat	α = linspace(−10,20,81)	linspace(−20,20,81)
		Width = linspace(2,30,11)
		γ = 0.02
		λ = 0.02
Adaptive Grid	Gaussian	Starting values for the adaptive grid:	linspace(−20,20,81)
		α = linspace(−20,20,81)
		β = linspace(0.2,2,11)
		γ = 0.02
		λ = 0.02
UML	Flat	α = linspace(−20,20,81)	[−20,20]
		β = logspace^c(0.2,2,11)
		γ = 0.02
		λ = linspace(0,0.2,5)

^a linspace(x₁, x₂, n) represents n linearly spaced values between x₁ and x₂.

^b The width of the PF was defined as the difference in SNR between the 5% and 95% correct intelligibility points on the unscaled PF (i.e. before the guessing and lapsing parameters are applied).

^c logspace(x₁, x₂, n) represents n logarithmically spaced values between x₁ and x₂.

Figure 1. Accuracy (absolute error in dB), bias (dB) and reliability (ICC) of each analysis-procedure combination for a target of SRT₅₀. Error bars indicate 95% confidence interval of the means.

Figure 2. Accuracy (absolute error in dB), bias (dB) and reliability (ICC) of each analysis-procedure combination for a target of SRT₇₅. Error bars indicate 95% confidence interval of the means.

Figure 3. Histograms of stimulus placements corresponding to individual listener’s intelligibilities with target intelligibilities of 50% (top row) and 75% (bottom row) respectively.

Figure 4. Accuracy (absolute error) and reliability (ICC) of each analysis-procedure combination for PF width estimation. Error bars indicate 95% confidence interval of the means.

Figure 5. Standardised regression coefficients for lapsing rate and threshold, target intelligibility, method and procedures. The ordinate indicates the predicted change in absolute error in dB for a one standardised unit change in the respective predictor. Error bars indicate the 95% confidence interval of the regression coefficient.