A WOLA-Based Real-Time Noise Reduction Algorithm to Improve Speech Perception with Cochlear Implants.

Background noise poses a significant challenge to people who have a cochlear implant for restoring their hearing ability. A cochlear implant can process sound into 12 to 24 channels and it provides limited temporal and spectral information to the auditory nerve through electrical current stimulation. In this paper, a specific noise reduction algorithm was developed to accommodate the need of adaptively applying a small number of gains to the stimulation signals in cochlear implants. A sound signal was first divided into 22 channels using the WOLA (Weighted Overlap Add) spectral analysis. The spectral templates of background noise were estimated by automatically tracking energy gaps between speech segments. The gap detection algorithm utilized a mechanism like the charging and discharging of a capacitor in an envelope detector, which offers the ability for the extracted energy signal to stay at noise floors. The noise templates were updated adaptively when a segment of signal was determined to be noise. Simulations of the proposed noise reduction algorithm were performed using offline processing and it has also been implemented on the Ezairo 7150 (ON Semiconductor Corporation) DSP platform with a WOLA co-processor. Initial evaluation results showed that the signal-to-noise (SNR) ratio can be improved by up to 10 dB after noise removal.