OPRA-RS: A Hearing-Aid Fitting Method Based on Automatic Speech Recognition and Random Search

Abstract

Hearing-aid (HA) prescription rules (such as NAL-NL2, DSL-v5, and CAM2) are used by HA audiologists to define initial HA settings (e.g., insertion gains, IGs) for patients. This initial fitting is later individually adjusted for each patient to improve clinical outcomes in terms of speech intelligibility and listening comfort. During this fine-tuning stage, speech-intelligibility tests are often carried out with the patient to assess the benefits associated with different HA settings. As these tests tend to be time-consuming and performance on them depends on the patient's level of fatigue and familiarity with the test material, only a limited number of HA settings can be explored. Consequently, it is likely that a suboptimal fitting is used for the patient. Recent studies have shown that automatic speech recognition (ASR) can be used to predict the effects of IGs on speech intelligibility for patients with age-related hearing loss (ARHL). The aim of the present study was to extend this approach by optimizing, in addition to IGs, compression thresholds (CTs). However, increasing the number of parameters to be fitted increases exponentially the number of configurations to be assessed. To limit the number of HA settings to be tested, three random-search (RS) genetic algorithms were used. The resulting new HA fitting method, combining ASR and RS, is referred to as "objective prescription rule based on ASR and random search" (OPRA-RS). Optimal HA settings were computed for 12 audiograms, representing average and individual audiometric profiles typical for various levels of ARHL severity, and associated ASR performances were compared to those obtained with the settings recommended by CAM2. Each RS algorithm was run twice to assess its reliability. For all RS algorithms, ASR scores obtained with OPRA-RS were significantly higher than those associated with CAM2. Each RS algorithm converged on similar optimal HA settings across repetitions. However, significant differences were observed between RS algorithms in terms of maximum ASR performance and processing costs. These promising results open the way to the use of ASR and RS algorithms for the fine-tuning of HAs with potential speech-intelligibility benefits for the patient.

Keywords: age-related hearing loss (ARHL); automatic speech recognition (ASR); compression thresholds; hearing aids (HAs); insertion gains; prescription rule; random search (RS).

Figure 1

Overview of the components of the OPRA-RS processing chain and associated output data. HA parameters randomized by the RS algorithm appear in red.

Figure 2

Schematic representation of GEN1, which simultaneously optimizes CTs and IGs in all HA channels.

Figure 3

Schematic representation of GEN2, which simultaneously optimizes the CT and IGs for one HA channel at the time, one after the other.

Figure 4

Schematic representation of GEN3, which optimizes first CTs and then IGs in all HA channels.

Figure 5

Individual and mean audiograms for different WARHICS levels used as input to the OPRA-RS processing chain (left panel). Pure-tone average (PTA) for frequencies between 0.5 to 4 kHz are shown for each audiogram in the right panel.

Figure 6

Distribution of ASR scores based on IGs selected by OPRA-RS or recommended by CAM2, for each of the three RS algorithms. The horizontal lines inside the boxes represent median ASR scores. Whiskers and horizontal limits of the boxes represent, from bottom to top, the 0, 25th, 75th, and 100th percentiles. In the case of OPRA-RS, the medians and 75th percentiles overlap.

Figure 7

IG functions prescribed by OPRA-RS and CAM2, averaged across audiograms.

Figure 8

Insertion gains prescribed by GEN1, GEN2, GEN3, and CAM2 using Humes (2021)'s average audiograms corresponding to WARHICS levels 4 to 7.

Figure 9

Distribution of the number of iterations needed to reach the highest ASR score that was reached by the three RS algorithms. Otherwise as in Figure 6.

Figure 10

IGs prescribed by OPRA-RS implementing each of the three RS algorithms (see rows) for Humes (2021)'s average audiograms corresponding to WARHICS levels 4 and 6 (see columns).

Figure 11

Absolute differences in CTs between repetitions for the three RS algorithms as a function of HA channel. Otherwise as in Figure 6.