Related Experiment Video
Updated: Jan 10, 2026
06:04
Systematic Hearing Performance Evaluation Process for Adolescents with Cochlear Implantation at Early Ages
Published on: March 24, 2023
736
Objective Evaluation of a Deep Learning-Based Noise Reduction Algorithm for Hearing Aids Under Diverse Fitting and
Vahid Ashkanichenarlogh1,2, Paula Folkeard1,3, Susan Scollie1,3
1National Centre for Audiology, Western University, London, Canada.
Trends in Hearing
|November 25, 2025
Summary
A deep-neural-network (DNN) combined with beamforming significantly improved speech intelligibility and sound quality in hearing aids. This advanced system outperformed traditional methods across various noise conditions and reverberation levels.
Area of Science:
- Auditory signal processing
- Artificial intelligence in acoustics
- Hearing aid technology
Background:
- Conventional hearing aid processing struggles with noise and reverberation.
- Deep learning offers potential for advanced noise reduction.
- Evaluating hearing aid performance requires robust intelligibility and quality metrics.
Purpose of the Study:
- To evaluate a deep-neural-network (DNN) denoising system integrated with beamforming.
- To compare the DNN system against adaptive filtering and beamforming alone.
- To assess the impact of noise types, SNRs, and hearing aid fittings on performance.
Main Methods:
- Utilized a KEMAR manikin with five audiograms in controlled acoustic environments.
- Generated 1,152 recordings across reverberant and non-reverberant conditions.
- Estimated speech intelligibility using the Hearing Aid Speech Perception Index (HASPI) and sound quality using HASQI and pMOS metrics.
Main Results:
- The DNN with beamforming demonstrated superior speech intelligibility compared to conventional methods.
- Performance gains were most pronounced at 0 and +5 dB SNR, with moderate benefits at -5 dB in low reverberation.
- Sound quality metrics (HASQI, pMOS) improved with SNR and showed moderate correlation, though pMOS exhibited greater variability.
Conclusions:
- Combining deep learning with beamforming offers significant benefits for hearing aid intelligibility and sound quality.
- Non-intrusive metrics like pMOS show promise for large-scale assessment but capture processing effects differently than intrusive metrics.
- The DNN-beamforming system shows potential for enhancing audiological rehabilitation, with performance modulated by acoustic environment and SNR.
