Electronics Letters (May 2024)

9.1 µW keyword spotting processor based on optimized MFCC and small‐footprint TENet in 28‐nm CMOS

  • Haohai Yu,
  • Keyan He,
  • Yang Liu,
  • Dihu Chen

DOI
https://doi.org/10.1049/ell2.13219
Journal volume & issue
Vol. 60, no. 9
pp. n/a – n/a

Abstract

Read online

Abstract This letter proposes a low‐power keyword spotting (KWS) architecture based on a modified temporal efficient neural network (TENet) and a simplified mel‐frequency cepstrum coefficient (MFCC) algorithm. The optimized MFCC algorithm reduces the computational load by 82% for multiplications and 66% for additions. An efficient hardware architecture and data flow for TENet have been designed, resulting in a 3.1× reduction in the operating cycle compared to similar work. The parameter count and computational load are reduced by 3.7× and 2.8×, respectively, and the accuracy reaches 95.36% for ten keywords in the Google Speech Command Dataset (GSCD). Operating at a frequency of 16 KHz for MFCC and 100 KHz for NN accelerator on a 28 nm process, the power consumption overhead is 9.1 µW.

Keywords