A Novel Artificial-Intelligence-Based Reverberation-Reduction Algorithm for Cochlear Implants Enhances Speech Intelligibility and User Experience.

The study tested participants at their individual RT60 50% threshold, which is a highly challenging condition. Benefits in typical daily environments with lower reverberation times may be smaller. The algorithms were trained on American English speech but tested on Dutch. While this study found no language barrier, tonal languages may pose a greater challenge. Testing involved simulated reverberation in a controlled setup. Real-life environments involve variable direct-to-reverberant ratios, source distances, and background noise sources (like competing talkers) which were not tested. The study measured acute effects using a research processor. Long-term adaptation and the impact of algorithmic latency (currently 40 msec) in daily use require further investigation. The sample size was small (n=15), limiting generalizability.

One author is a member of the European Medical Advisory Board of Advanced Bionics, the company whose processor was used for testing. The study was funded by Stichting voor de Technische Wetenschappen.

Do not assume these algorithms are currently available for clinical use. This was a lab-based study using a research processor and simulated reverberation. Real-world performance in complex acoustic environments with multiple noise sources, and the effects of long-term use and algorithmic latency, have not yet been established.