Related Experiment Video
Updated: Sep 28, 2025
06:04
Systematic Hearing Performance Evaluation Process for Adolescents with Cochlear Implantation at Early Ages
Published on: March 24, 2023
484
Using Automatic Speech Recognition to Optimize Hearing-Aid Time Constants
Lionel Fontan1, Libio Gonçalves Braz2, Julien Pinquier2
1Archean LABS, Montauban, France.
Frontiers in Neuroscience
|April 4, 2022
Summary
Optimizing hearing aid (HA) time constants significantly improved speech recognition for age-related hearing loss when using standard gains. However, optimization did not benefit ASR-predicted performance with advanced, ASR-derived gains.
Area of Science:
- Audiology
- Speech Processing
- Computational Auditory Perception
Background:
- Automatic speech recognition (ASR) combined with hearing loss (HL) and hearing aid (HA) simulations can predict speech identification performance in individuals with age-related hearing loss.
- ASR facilitates the evaluation and optimization of HA configurations, including insertion gains and compression thresholds, for personalized HA fitting.
Purpose of the Study:
- To investigate the efficacy of a random-search algorithm in optimizing hearing aid time constants (attack and release times) for 12 distinct audiometric profiles.
- To compare the impact of optimizing time constants using standard CAM2 insertion gains versus ASR-optimized gains, with ASR-optimized compression thresholds.
Main Methods:
- A hearing aid simulator and a hearing loss simulator were employed to process speech stimuli based on audiometric profiles.
- A random-search algorithm was utilized to iteratively adjust time constants, aiming to maximize ASR performance for each audiometric profile over 1,000 iterations.
- The random search was performed twice to evaluate the reproducibility of the optimized time-constant configurations and resulting ASR scores.
Main Results:
- Optimizing time constants led to a significant improvement in ASR scores when using CAM2 insertion gains.
- No significant improvement in ASR scores was observed when optimizing time constants with ASR-based insertion gains.
- While repeating the random search produced similar ASR scores, the specific time-constant configurations varied.
Conclusions:
- Time constant optimization is beneficial for improving ASR-predicted speech recognition with conventional HA gain prescriptions.
- The benefits of time constant optimization may be limited when using advanced, ASR-derived gain strategies.
- Further research is needed to understand the interaction between different HA parameters and their impact on speech recognition outcomes.
