Evaluation of a sparse coding shrinkage algorithm in normal hearing and hearing impaired listeners

Hearing impaired (HI) people struggle more than normal hearing (NH) listeners to understand speech in noisy environment. Previous evaluations of noise reduction algorithms on HI listeners have mainly concentrated on few algorithms like spectral subtraction or Wiener filtering. In this paper, a sparse coding shrinkage (SCS) noise reduction algorithm is proposed to compensate for some of the auditory deficits. The noise reduction performance by the SCS algorithm is compared with a Wiener filtering (CS-WF) approach, where the a priori signal-to-noise-ratio is estimated by the cepstral smoothing method. Speech recognition tests were performed to assess subjective intelligibility of SCS, CS-WF and noisy speech in babble noise and speech-shaped noise. Results show that both noise reduction algorithms have more potential to improve speech intelligibility in HI listeners than NH listeners; SCS provides more benefits than CS-WF for HI listeners especially in speech shaped noise.