Speech Enhancement Based on Harmonic Estimation Combined with MMSE to Improve Speech Intelligibility for Cochlear Implant Recipients
Abstract
In this paper, a speech enhancement algorithm is proposed to improve the speech intelligibility for cochlear implant recipients. Our method is based on combination of harmonic estimation and traditional statistical method. Traditional statistical based speech enhancement method is effective only for stationary noise suppression, but not non-stationary noise. To address more complex noise scenarios, we explore the harmonic structure of target speech to obtain a more accurate noise estimation. The estimated noise is then employed in the MMSE framework to obtain the gain function for recovering the target speech. Listening test experiments show a substantial speech intelligibility improvement for cochlear implant recipients in noisy environments.