Improving the performance of hearing aids in noisy environments based on deep learning technology

Area of Science:

• Audiology
• Artificial Intelligence
• Signal Processing

Background:

• Hearing aid users often struggle with speech perception in noisy environments.
• Existing speech enhancement (SE) technologies have limitations in restoring speech intelligibility and quality.

Purpose of the Study:

• To evaluate the performance of a deep-learning-based SE technology, the deep denoising autoencoder (DDAE), for hearing aid users.
• To compare the DDAE approach against classical SE methods using standard evaluation metrics.

Main Methods:

• The DDAE SE approach was tested on individuals with typical high-frequency hearing loss (HFHL) audiograms.
• Performance was assessed using the hearing-aid speech perception index (HASPI) for intelligibility and the hearing-aid sound quality index (HASQI) for quality.

Main Results:

• The DDAE SE approach demonstrated superior performance compared to two classical SE methods.
• Higher scores for both speech intelligibility (HASPI) and quality (HASQI) were achieved with the DDAE method.

Conclusions:

• Deep-learning-based SE, specifically the DDAE method, shows significant potential for enhancing hearing aid user experience.
• The DDAE technology can effectively improve speech intelligibility and quality in noisy conditions for individuals with hearing loss.