A Spectral Change Enhancement Method Based on Self-Supervised Learning Framework

Modern hearing aids often fail to benefit the wearer in noisy environments. In fact, the auditory processing of hearing impaired listeners usually leads to spectral smearing, and hearing aids compression processing furtherly decreases the spectral and temporal contrasts of incoming sound, which both influence the perception of speech in background noise for the hearing impaired listeners. To solve this problem, this paper proposes a simple but effective enhancement neural network based on self-supervised learning framework to enhance the spectral contrast of the noisy speech. Both objective evaluation and the result of subjective experiments indicate that our method can improve speech perception of listeners with reduced frequency selectivity of auditory system.