To demonstrate that affordable, offline sound-awareness technology for deaf and hard-of-hearing (DHH) users is achievable in low-income settings using edge AI.
Design Science Research and Tankyu Practice methodologies were applied in Benin. Four context-specific TinyML models were deployed on a USD 30 microcontroller-based device and evaluated with 26 DHH participants over 40 days.
All four models achieved 90.7- 97.9% laboratory accuracy. On-device offline inference delivered a latency of 267 ms, within the 500 ms alert threshold. The mean SUS score was 72.4.
This is the first evaluation of a DHH sound-awareness wearable with participants in sub-Saharan Africa, with a community contribution model that inverts the recording burden of prior personalisation systems.
The research is based on a single-city sample. Field accuracy was not quantitatively measured. Extension to other LMIC contexts is needed.
A USD 30 offline device is technically feasible as a research prototype. Systematic data collection and false positive mitigation are required before production deployment.
Noutchogbe, C. A. A. (2026): EchoBand and Pulses: Developing AI Solutions for Sustainable Sound Accessibility in Low-Income Settings. World Journal of Entrepreneurship, Management and Sustainable Development (WJEMSD), Vol. 22, No. 5, pp. 403-418.
