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

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...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Electronics Letters 2024-05, Vol.60 (9), p.n/a
Hauptverfasser: Yu, Haohai, He, Keyan, Liu, Yang, Chen, Dihu
Format: Artikel
Sprache:eng
Schlagworte:
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung: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. 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. Operating at a frequency of 16 KHz for MFCC and 100 KHz for NN accelerator on a 28nm process, the power consumption overhead is 9.1 microwatts, and the accuracy reaches 95.36% for 10 keywords in the Google Speech Command Dataset (GSCD).
ISSN:0013-5194
1350-911X
DOI:10.1049/ell2.13219