Download or read book Investigations on the Deep Learning Based Speech Enhancement Algorithms for Hearing-impaired Population written by Zhuohuang Zhang and published by . This book was released on 2022 with total page 0 pages. Available in PDF, EPUB and Kindle. Book excerpt: Speech signals are often contaminated by unwanted background noise or interfering speakers that makes daily communication more difficult, especially for hearing-impaired (HI) listeners. Speech enhancement algorithms are thus proposed to alleviate this problem, by removing unwanted sounds. A plethora of these algorithms have been proposed over the years, however, most of them have been optimized and evaluated using only objective measures due to the efficiency of these measures and the cost of alternative methods. These objective measures may not generalize well with subjective responses. Moreover, the underlying mechanisms of the human auditory system for speech quality judgment is not well understood, further investigations are needed.It is important to investigate the factors that contribute to speech quality perceived by human listeners, to facilitate future development of speech enhancement systems with improved efficiency and user experience. There are several research questions that are worth investigating on the design of speech enhancement algorithms, including (1) How do existing speech enhancement algorithms perform on HI listeners? (2) Could HI listeners perceive the phase distortion contained in a speech signal? And (3) Does phase estimation at different frequency bands contribute equally to perceived speech quality?Contributions of this dissertation involve the following three aspects. First, we thoroughly investigated the performance of certain speech enhancement algorithms for both simulated HI and normal hearing (NH) listeners. Second, we conducted a series of listening studies to investigate how human perception is impacted by phase distortions in speech signals. Third, we inspected the band importance of estimating phase on speech quality through several pairwise comparison experiments. We further proposed a novel hybrid speech enhancement framework to efficiently perform phase-aware noise reduction.