Although hearing aids can make everyday life easier, many people with hearing loss still choose not to use them because they fear the stigma often associated with wearing them. “Younger people, in particular, often feel uncomfortable because hearing aids are commonly associated with old age,” explains Vaquero. In their Bachelor’s thesis, the two students therefore sought solutions to make hearing aids less conspicuous when worn. By developing a system capable of recognising gestures, they focused primarily on creating a more discreet mode of operation. “This means you can operate the hearing aid with a subtle nod or a shake of the head,” explains Vaquero. To better understand users’ needs, they interviewed people who wear hearing aids.
Daniela Monteiro
“Through this work, we were able to make a small contribution towards potentially making life easier for some people. That’s very rewarding.”
Goal: two seconds
“The first crucial question we wanted to clarify was: How quickly does the system need to respond in order to be practical for everyday use?” says Monteiro. It was clear to all those surveyed that a hearing aid cannot (yet) respond to a nod as quickly as a smartphone responds to a tap on a touchscreen. But how much time is too much? Drawing on their interviews and academic literature, the graduates set themselves a target response time of two seconds. For gesture recognition, they tested two different approaches: a machine learning method in which the system was trained to recognise gestures and a dynamic time warping algorithm as used in speech recognition. They then developed two approaches: a machine learning method, in which the system was trained to recognise gestures, and an algorithm for time alignment (Dynamic Time Warping), as used in speech recognition.
Nino Hail
Senior Software Engineer, Sonova Group R&D
“Carla and Daniela’s work lays valuable foundations and demonstrates promising approaches for future technological developments.”
The model will continue to be developed
One unexpected challenge was the degree of variation in human gestures. “Of course, we were aware that gestures vary across cultures and generations. However, the fact that there are such significant differences even within the same group posed a real conundrum for our project,” says Vaquero. Monteiro adds: “In the end, we simply had to accept certain limits. Given the size of the device and the limited time available before submission, we had to restrict ourselves to a simple model.” Her colleague continues: “We would have liked to have researched the topic in even greater depth. Within the scope of our Bachelor’s thesis, however, we did not have the necessary resources.” Even so, they succeeded in developing a small-scale model that already performed remarkably well. “It provides a foundation that can be built upon,” says Vaquero. This is precisely what will happen now. Their industry partner Sonova, where Vaquero works, will take over the code and continue its development.
