HEARING LOSS, SPECTRAL DEFICITS AND FORMANT ENHANCEMENT

点击翻译按钮获取中文摘要

The overall goal of this project is to evaluate promising signal processing techniques to enhance speech spectral characteristics for improved speech understanding by hard-of-hearing (HoH) listeners. These spectral enhancements include: 1) deepening of between-formant spectral valleys, 2) frequency scaling of speech, and 3) combinations of both deepening and scaling. Many hard-of- hearing (HoH) listeners can resolve low-frequency spectral cues normally but they are often less able to resolve frequency information in higher-frequency signals. In speech, higher- frequency second formants (F2), which provide important information for phoneme distinctions, are not easily resolved by these HoH listeners. Sinusoidal modeling techniques will be used to detect spectral peaks in rapidly changing syllables, and to deepen spectral valleys between peaks. Sound design software will be used to accomplish modest frequency scaling (up to 30 percent reduction) while retaining good sound quality. It is anticipated that these combined schemes will produce good-quality speech signals with formants that are more easily resolved by HoH listeners. Two groups of listeners with moderate to severe hearing loss will be identified as having better or poorer spectral resolution abilities based on frequency discrimination and spectral peak detection tasks. These listeners will then identify unmodified stimuli and stimuli that have been enhanced by the proposed processing chemes. Results will demonstrate the benefits of each scheme individually, and of the combined schemes. Early results using pilot processed stimulil show promise toward the goal of improving speech perception abilities of listeners with moderate to severe hearing loss. Results from the proposed investigation will provide important additional pilot data that will serve as a basis for future investigations of speech enhancements. A long-term goal of this research is the development of real-time enhancement algorithms that can be implemented into wearable amplification devices.