Manual operation of hearing assistive devices is cumbersome in a number of situations. To assist in addressing this issue, voice interfaces are envisioned as a means for handling and operating hearing assistive devices in a practical manner. Furthermore, it is key that such voice interfaces take into account that hearing assistive devices are characterized by strict memory and computational complexity constraints.
In spite all the progress made in both machine learning and speech technology in recent years, there is still a long way to go in the development of voice interfaces that operate flawlessly in acoustically challenging (i.e., noisy) situations. Therefore, the goal of this project is the research and development of personalized, noise-robust and low-resource keyword spotting systems for hearing assistive devices. To meet all these requirements, we explore the combined use of multi-microphone signals from hearing assistive devices along with signal processing and the latest deep learning techniques. We expect to improve the robustness and performance over existing keyword spotting systems by exploring systems that take into account user-specific aspects, e.g., voice characteristics or head-related acoustics of the specific user. As a result, we also expect to contribute to enhance the life quality of hearing-impaired people.
