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

Area of Science:

• Audiology
• Artificial Intelligence
• Signal Processing

Background:

• Hearing aid (HA) prescription rules (e.g., NAL-NL2, DSL-v5, CAM2) guide initial HA settings.
• Current fine-tuning relies on time-consuming speech intelligibility tests, potentially leading to suboptimal patient fittings.
• Automatic speech recognition (ASR) shows promise in predicting the impact of insertion gains (IGs) on speech intelligibility for age-related hearing loss (ARHL).

Purpose of the Study:

• To extend ASR applications by optimizing both IGs and compression thresholds (CTs) for hearing aids.
• To develop an objective hearing aid fitting method combining ASR with random-search (RS) genetic algorithms.
• To compare the performance of the new method (OPRA-RS) against a standard prescription rule (CAM2).

Main Methods:

• Developed an objective prescription rule based on ASR and random search (OPRA-RS).
• Utilized three RS genetic algorithms to limit the number of configurations assessed for optimizing IGs and CTs.
• Computed optimal HA settings for 12 audiograms representing various ARHL severities and compared ASR performance with CAM2 settings.

Main Results:

• OPRA-RS achieved significantly higher ASR scores compared to the CAM2 rule across all RS algorithms.
• RS algorithms demonstrated reliability, converging on similar optimal HA settings across repetitions.
• Differences were noted between RS algorithms regarding maximum ASR performance and computational cost.

Conclusions:

• The OPRA-RS method, combining ASR and RS, offers a promising approach for hearing aid fine-tuning.
• This objective method has the potential to enhance speech intelligibility for patients with hearing loss.
• Further research can explore the clinical benefits and refine the application of ASR and RS in audiology.