Neu-BrAuER: a neuromorphic Braille letters audio-reader for commercial edge devices

A widespread deployment of neuromorphic models on edge devices is still significantly hindered by the rarity of dedicated hardware. Despite the more and more relevant results achieved working with neuromorphic chips like Intel’s Loihi, the vast majority of commercially available edge devices still do not employ specific architectures intended to emulate brain-like computation. As a result, investigation of solutions for neuromorphic applications that leave space for deployment on more common edge devices like, for instance, traditional Microprocessor Units (MPUs) is needed. Only with further exploration in this direction is indeed possible to foster the adoption of Spiking Neural Networks (SNNs) for on-edge deployment of neuromorphic models. To this aim, we show the Neuromorphic Braille Audio-Enhanced Reader (Neu-BrAuER), designed to perform classification and audio-reading of Braille letters on edge devices with traditional architectures. Furthermore, Neu-BrAuER does not require previous spike encoding, thus allowing to directly feed data from off-the-shelf sensors to the neuromorphic classifier. By showing the suitability of SNN-based models for on-edge and real-time tasks, Neu-BrAuER can represent a seed for other brain-inspired solutions in different domains that require, regardless of the available hardware, to deal with time-varying signals adopting a neuromorphic approach.