Speech Enhancement for Degraded Radio Speech
Akio Kobayashi, Keiichi Yasu · 2024
In Japan, radio broadcasting is regarded as an important disaster medium; however, it can be extremely difficult to listen to, not only for hard-of-hearing people but also for normal-hearing people, as it is susceptible to interference, such as noise and fading, which can degrade the quality of the received speech signals. Low-quality radio speech can be improved using speech enhancement techniques; however, whether robust model training can be achieved in different reception environments has rarely been explored. Therefore, this study investigates the effectiveness of speech enhancement for radio speech from the models trained on data collected at different reception points and pseudo-generated training data.