A Separable Temporal Convolution Neural Network with Attention for Small-Footprint Keyword Spotting

TL;DR

This paper introduces a lightweight separable temporal convolution neural network with attention for keyword spotting on mobile devices, achieving high accuracy with significantly fewer parameters than existing models.

Contribution

The paper presents a novel small-footprint model combining separable temporal convolutions and attention, maintaining high performance with only 32.2K parameters.

Findings

Achieves 95.7% accuracy on Google Speech Commands dataset

Uses only 32.2K parameters, much fewer than state-of-the-art models

Maintains high performance with a small model size

Abstract

Keyword spotting (KWS) on mobile devices generally requires a small memory footprint. However, most current models still maintain a large number of parameters in order to ensure good performance. To solve this problem, this paper proposes a separable temporal convolution neural network with attention, it has a small number of parameters. Through the time convolution combined with attention mechanism, a small number of parameters model (32.2K) is implemented while maintaining high performance. The proposed model achieves 95.7% accuracy on the Google Speech Commands dataset, which is close to the performance of Res15(239K), the state-of-the-art model in KWS at present.