Preprint
Hello Edge: Keyword Spotting on Microcontrollers
Yundong Zhang et al., arXiv, 2017
Read the originalRead on October 17, 2026
- What it is
- A paper from Arm researchers that designs neural networks for keyword spotting, which detects a small set of spoken keywords, under the memory and power limits of microcontrollers. It measures accuracy against model size on those constraints.
- How much weight it carries
- A preprint, posted to arXiv and revised in 2018. It studies keyword spotting, not open-vocabulary dictation, and it targets the chips of its time. Newer chips and newer models will differ.
Facts taken from this source
- The Hello Edge authors say keyword spotting, which must always listen, typically runs on tiny microcontrollers with limited memory and compute capability, so the network design has to fit those limits.
“typically runs on tiny microcontrollers with limited memory and compute capability” (Abstract)