Speech enhancement for cochlear implant recipients.

Area of Science:

• Signal Processing
• Auditory Neuroscience
• Biomedical Engineering

Background:

• Cochlear implant recipients often struggle with speech intelligibility in noisy environments.
• Traditional speech enhancement methods, like Minimum Mean Square Error (MMSE), are less effective with non-stationary noise.
• Existing algorithms may not fully address the complex acoustic challenges faced by cochlear implant users.

Purpose of the Study:

• To propose a novel single-microphone speech enhancement algorithm.
• To improve speech intelligibility for cochlear implant recipients in various noise conditions.
• To address limitations of traditional methods in handling non-stationary noise.

Main Methods:

• The algorithm combines harmonic structure estimation with a statistical speech enhancement stage.
• It estimates noise structure in both time and frequency dimensions, exploiting speech harmonics.
• Noise estimation is integrated into a Minimum Mean Square Error (MMSE) framework.

Main Results:

• A formal listener evaluation was conducted with 6 cochlear implant recipients.
• The proposed algorithm demonstrated substantial improvements in speech intelligibility.
• The method effectively enhances speech in the presence of both stationary and non-stationary noise.

Conclusions:

• The developed speech enhancement algorithm significantly benefits cochlear implant users.
• Exploiting harmonic structure enhances noise estimation for better speech enhancement.
• This approach offers a promising solution for improving auditory perception in challenging acoustic environments.