BPACE: A Bayesian, Patient-Centered Procedure for Matrix Speech Tests in Noise

Abstract

Matrix sentence tests in noise can be challenging to the listener and time-consuming. A trade-off should be found between testing time, listener's comfort and the precision of the results. Here, a novel test procedure based on an updated maximum likelihood method was developed and implemented in a German matrix sentence test. It determines the parameters of the psychometric function (threshold, slope, and lapse-rate) without constantly challenging the listener at the intelligibility threshold. A so-called "credible interval" was used as a mid-run estimate of reliability and can be used as a termination criterion for the test. The procedure was evaluated and compared to a STAIRCASE procedure in a study with 20 cochlear implant patients and 20 normal hearing participants. The proposed procedure offers comparable accuracy and reliability to the reference method, but with a lower listening effort, as rated by the listeners ($-1.8$ points on a 10-point scale). Test duration can be reduced by 1.3 min on average when a credible interval of 2 dB is used as the termination criterion instead of testing 30 sentences. Particularly, normal hearing listeners and well performing, cochlear implant users can benefit from shorter test duration. Although the novel procedure was developed for a German test, it can easily be applied to tests in any other language.

Keywords: adaptive procedure; listening effort; psychometric function; signal-to-noise ratio; speech in noise.

Figure 1.

Listener’s responses (black dots) depending on the signal-to-noise ratio (SNR) with the resulting psychometric function and its sweet-points ( $\alpha$ as triangle, $\beta$ as squares and $\lambda$ as diamond).

Figure 2.

Timeline of the study.

Figure 3.

SRT for the STAIRCASE and the BPACE methods separated for CI patients and NH listeners. Abbreviations: SRT= speech reception threshold; CI = cochlear implant; NH = normal hearing; BPACE = Bayesian PAtient-CEntered.

Figure 4.

SRT depending on test sequence number for CI patients and NH listeners. Abbreviations: SRTs = speech reception thresholds; CI = cochlear implant; NH = normal hearing.

Figure 5.

Left: Bland–Altman plot for STAIRCASE and BPACE test methods. Right: Scatterplot of the SRT, the color separates NH listeners from CI patients. Abbreviations: SRTs = speech reception thresholds; CI = cochlear implant; NH = normal hearing.

Figure 6.

Subjective listening effort and mean proportion correct in the adaptive test for the CI patients and the NH participants. Abbreviations: CI = cochlear implant; NH = normal hearing.

Figure 7.

BPACE test duration for 30 trials or until reaching the desired credible interval for the CI patients and NH participants. The dashed line indicates the mean duration of the STAIRCASE method (4.8 min). Abbreviations: BPACE = Bayesian PAtient-CEntered; CI = cochlear implant; NH = normal hearing.