EchoBand and pulses: developing AI solutions for sustainable sound accessibility in low-income settings, Cyrille Noutchogbe

Cyrille Noutchogbe
Kobe Institute of Computing
Japan

Purpose: To show that human-centered design combined with edge AI can produce affordable, offline assistive technology for deaf and hard-of-hearing (DHH) individuals in low-income settings where commercial solutions cost USD 200-500.
Design/methodology/approach: Design Science Research was conducted in Benin, West Africa, using the Tankyu Practice methodology. Four context-specific TinyML models were trained and deployed on an M5StickC Plus 2 microcontroller. Field evaluation involved 26 DHH participants at CAEIS in Porto-Novo over 40 days.
Findings: Laboratory validation accuracy ranged from 90.7% to 97.9% across four context models. End-to-end latency averaged 267 milliseconds. The mean System Usability Scale score was 72.4 (Good). Participants confirmed that bimodal haptic-plus-visual feedback matters even when no alternatives exist.
Original/value of the paper: This is the first evaluation of a DHH sound-awareness wearable with participants in sub-Saharan Africa. The community-based data contribution model inverts the individual-recording burden of all prior personalization systems.
Research limitations/implications: The sample covers one West African city. The formal field accuracy measurement was not conducted; only a qualitative observation was conducted. Future work should extend to rural settings and other francophone urban contexts.
Practical implications: A USD 30 offline device is technically viable. Cost remains a barrier even at this level, suggesting that subsidy mechanisms or community ownership models deserve attention.
Keywords: Assistive Technology; Deaf and Hard-of-Hearing; TinyML; Sustainable Development; Edge AI; Low-income Contexts; Community-based Innovation; Wearable Devices; West Africa; Benin