How Tiny Can Analog Filterbank Features Be Made for Ultra-low-power On-device Keyword Spotting?

TL;DR

This paper investigates how to minimize the size of analog filterbank features for ultra-low-power on-device keyword spotting, demonstrating significant power reduction with minimal accuracy loss.

Contribution

It presents a simulation-based analysis showing that careful parameter tuning can drastically reduce analog filterbank power consumption with minimal impact on accuracy.

Findings

Power can be reduced by 33.6x through parameter optimization.

Minimal accuracy loss of 1.8% in keyword spotting.

Analog filterbank parameters critically influence power efficiency.

Abstract

Analog feature extraction is a power-efficient and re-emerging signal processing paradigm for implementing the front-end feature extractor in on device keyword-spotting systems. Despite its power efficiency and re-emergence, there is little consensus on what values the architectural parameters of its critical block, the analog filterbank, should be set to, even though they strongly influence power consumption. Towards building consensus and approaching fundamental power consumption limits, we find via simulation that through careful selection of its architectural parameters, the power of a typical state-of-the-art analog filterbank could be reduced by 33.6x, while sacrificing only 1.8% in downstream 10-word keyword spotting accuracy through a back-end neural network.