Encoder-Decoder Neural Architecture Optimization for Keyword Spotting

TL;DR

This paper introduces a neural architecture search method using encoder-decoder optimization to design specialized CNN models for keyword spotting, achieving state-of-the-art accuracy with efficient memory use.

Contribution

It presents a novel neural architecture search approach tailored for keyword spotting, outperforming traditional off-the-shelf models.

Findings

Achieves over 97% accuracy on Speech Commands Dataset.

Designs optimized CNN architectures outperform standard backbones.

Maintains acceptable memory footprint for practical deployment.

Abstract

Keyword spotting aims to identify specific keyword audio utterances. In recent years, deep convolutional neural networks have been widely utilized in keyword spotting systems. However, their model architectures are mainly based on off-the shelfbackbones such as VGG-Net or ResNet, instead of specially designed for the task. In this paper, we utilize neural architecture search to design convolutional neural network models that can boost the performance of keyword spotting while maintaining an acceptable memory footprint. Specifically, we search the model operators and their connections in a specific search space with Encoder-Decoder neural architecture optimization. Extensive evaluations on Google's Speech Commands Dataset show that the model architecture searched by our approach achieves a state-of-the-art accuracy of over 97%.