A Configurable Accelerator for Keyword Spotting Based on Small-Footprint Temporal Efficient Neural Network

Keyword spotting (KWS) plays a crucial role in human–machine interactions involving smart devices. In recent years, temporal convolutional networks (TCNs) have performed outstandingly with less computational complexity, comparison classical neural network (CNN) methods. However, it remains challenging to achieve trade-off between small-footprint model and high accuracy for the edge deployment of KWS system. this article, we propose based on modified efficient (TENet) simplified mel-frequency cepstrum coefficient (MFCC) algorithm. With batch-norm folding int8 quantization network, our achieves 95.36% Google Speech Command Dataset (GSCD) only 18 K parameters 461 multiplications. Furthermore, following hardware/model co-design approach, an optimized dataflow configurable hardware architecture TENet inference. The proposed accelerator implemented Xilinx zynq 7z020 energy efficiency 25.6 GOPS/W reduces runtime by 3.1× compared state-of-the-art work.