A Fused Deep Denoising Sound Coding Strategy for Bilateral Cochlear Implants.

Area of Science:

• Audiology and Speech-Language Pathology
• Biomedical Engineering
• Artificial Intelligence in Healthcare

Background:

• Severe sensorineural hearing loss impacts speech comprehension, especially in noise.
• Bilateral cochlear implants (BiCI) offer improved hearing but face challenges in noisy environments.
• Advanced signal processing, including deep learning, is crucial for effective noise reduction in CI users.

Purpose of the Study:

• To develop and evaluate a deep-learning-based bilateral speech enhancement model for BiCI systems.
• To improve spatial hearing and noise reduction capabilities in individuals with bilateral severe sensorineural hearing loss.
• To investigate the effectiveness of a novel latent fusion strategy for interaural information sharing.

Main Methods:

• Proposed a bilateral speech enhancement model integrating two monaural end-to-end deep denoising sound coding techniques.
• Utilized intermediary latent fusion layers to combine representations from both hearing sides via multiplication.
• Evaluated the model using objective instrumental measures and subjective speech-in-noise intelligibility tests with BiCI users.

Main Results:

• The fused BiCI sound coding strategy demonstrated higher interaural coherence and superior noise reduction compared to baseline methods.
• Objective measures indicated enhanced predicted speech intelligibility.
• BiCI users achieved speech-in-noise intelligibility scores comparable to quiet listening conditions.

Conclusions:

• The proposed deep denoising sound coding strategy effectively enhances speech intelligibility in noisy environments for BiCI users.
• Latent fusion in BiCI systems offers a promising approach for improving hearing aid performance.
• This technology has the potential to significantly improve the quality of life for individuals with bilateral hearing loss.