Abstract:This paper proposes a high power-performance-area efficient background noise aware keyword-spotting (KWS) processor based on an optimized binarized weight network (BWN). To reduce the power consumption while maintaining the system recognition accuracy for different background noise, the KWS processor with a SNR prediction module can be adaptively configured to use dual computing modes (standard computing mode and approximate computing mode) for both high recognition accuracy under high background noise and ultra-low power consumption under low background noise. The mel-scale frequency cepstral coefficients (MFCC) module is optimized with approximate computing technologies, which can reduce the power consumption by up to <inline-formula> <tex-math notation="LaTeX">$3.1\times $ </tex-math></inline-formula> and <inline-formula> <tex-math notation="LaTeX">$5.7\times $ </tex-math></inline-formula> for high/low background noise, respectively. Based on the evaluation of the architecture design space exploration, an ultra-low power BWN accelerator with low voltage, area and leakage power and using precision self-adaptive approximate computing units was proposed. Evaluated under 22nm process technology, this work can support up to 10 keywords real time recognition with power consumption of <inline-formula> <tex-math notation="LaTeX">$15.1~\mu \text{W}$ </tex-math></inline-formula> for high background noise and <inline-formula> <tex-math notation="LaTeX">$10.8~\mu \text{W}$ </tex-math></inline-formula> for low background noise. Compared to the state-of-the-art KWS architectures, our work can achieve ultra-low power consumption (about <inline-formula> <tex-math notation="LaTeX">$1.7\times $ </tex-math></inline-formula> reduced), while maintaining high system capability and adaptability.

Energy-Friendly Keyword Spotting System Using Add-Based Convolution.

A Depthwise Separable Convolution Neural Network for Small-footprint Keyword Spotting Using Approximate MAC Unit and Streaming Convolution Reuse

NS-KWS: joint optimization of near-sensor processing architecture and low-precision GRU for always-on keyword spotting

Separable Temporal Convolution plus Temporally Pooled Attention for Lightweight High-performance Keyword Spotting

Sub-mW Keyword Spotting on an MCU: Analog Binary Feature Extraction and Binary Neural Networks

A 510-nW Wake-Up Keyword-Spotting Chip Using Serial-FFT-Based MFCC and Binarized Depthwise Separable CNN in 28-nm CMOS

14.1 A 510nw 0.41V Low-Memory Low-Computation Keyword-Spotting Chip Using Serial FFT-Based MFCC and Binarized Depthwise Separable Convolutional Neural Network in 28nm CMOS

Global-Local Convolution with Spiking Neural Networks for Energy-efficient Keyword Spotting

Effective Combination of DenseNet and BiLSTM for Keyword Spotting.

Low Power Keyword Recognition Accelerator based on Approximate Calculation of Deep-Shift Neural Network

Text Adaptive Detection for Customizable Keyword Spotting.

NS-FDN: Near-Sensor Processing Architecture of Feature-Configurable Distributed Network for Beyond-Real-Time Always-on Keyword Spotting

End-to-end keyword search system based on attention mechanism and energy scorer for low resource languages

Hello Edge: Keyword Spotting on Microcontrollers

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

Small-Footprint Keyword Spotting with Multi-Scale Temporal Convolution

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

A Low-power and High-accuracy Accelerator with Voice Classification for Keyword Spotting

Focal Loss And Double-Edge-Triggered Detector For Robust Small-Footprint Keyword Spotting

TDT-KWS: Fast And Accurate Keyword Spotting Using Token-and-duration Transducer

A 22nm, 10.8 μ W/15.1 μ W Dual Computing Modes High Power-Performance-Area Efficiency Domained Background Noise Aware Keyword- Spotting Processor