Rubbing-noise-reduction Approach for Body-worn Hearing-aid Users Based On
deep-learning Technology
Presentation Number:0019 Time:15:00 - 15:12
|
Sheng-Jun Zhong, Pei-Chun Li, Woei-Chyn Chu, Shih-Tsang Tang, Sin-Hua Jhang and Ying-Hui Lai
Body-worn hearing aids are one of the most common assistive-hearing devices for individuals with hearing difficulties. It mainly integrates microphones, speakers, and signal-processing units to enhance the acoustic signals thereby improving the audibility for hearing-impaired individuals. Previous studies have shown that the pocket aid benefits the users; however, there is still room for improvement, such as reduction of rubbing noise. Mechanism design and application of acoustic absorbing damping material are common approaches for solving this issue; however, such approaches offer limited improvement. Therefore, a suitable signal-processing method, such as the deep-denoising autoencoder (DDAE), could be used to further reduce the rubbing noise. The experimental results show that the DDAE provides higher perceptual evaluation of speech quality and extended short-time objective intelligibility than the classical noise reduction (NR) approach. Moreover, multi-objective learning-based DDAE can achieve higher performance than DDAE NR. These results suggest that the deep-learning-based NR could be used to further improve the benefits for body-worn hearing aid users.
|
|
|
|
|
|
|
|