A Novel Artificial-Intelligence-Based Reverberation-Reduction Algorithm for Cochlear Implants Enhances Speech...

Area of Science:

• Audiology
• Artificial Intelligence
• Signal Processing

Background:

• Cochlear implants (CIs) are standard for severe-to-profound hearing loss.
• Speech understanding in reverberant environments remains a challenge for CI users.
• Current CI technology struggles with background noise and echo.

Purpose of the Study:

• To evaluate novel AI-based algorithms for reducing reverberation in speech signals for CI users.
• To assess the impact of these algorithms on speech intelligibility and subjective listening experience.
• To determine if AI dereverberation can improve CI performance in challenging acoustic conditions.

Main Methods:

• A prospective crossover study involving 15 CI users.
• Tested two AI algorithms: DNN-WPE (late reverberation) and DNN-WPEPF (early and late reverberation).
• Speech intelligibility measured using the Flemish/Dutch Matrix test; subjective ratings assessed listening effort, naturalness, and intelligibility.

Main Results:

• DNN-WPE improved speech intelligibility by 11% (p < 0.001); DNN-WPEPF improved it by 17% (p < 0.001).
• DNN-WPEPF showed significantly greater benefit than DNN-WPE (p = 0.018).
• Both algorithms were preferred for listening effort, naturalness, and intelligibility in reverberant conditions and did not degrade clean speech.

Conclusions:

• AI-driven dereverberation algorithms (DNN-WPE, DNN-WPEPF) significantly benefit CI users in reverberant environments.
• These algorithms enhance speech intelligibility and subjective perception without compromising performance in quiet.
• Further research and real-time implementation are needed for widespread clinical use.