Improving vision-inspired keyword spotting using dynamic module skipping in streaming conformer encoder

TL;DR

This paper introduces a dynamic module skipping mechanism in a vision-inspired streaming conformer encoder for keyword spotting, significantly reducing processing on non-speech inputs while maintaining high accuracy.

Contribution

It presents a novel architecture with trainable binary gates enabling input-dependent dynamic depth in a conformer encoder for streaming keyword spotting.

Findings

Achieves improved detection and localization accuracy on Librispeech.

Reduces average processing by up to 97% on non-speech inputs.

Maintains performance while significantly lowering computational load.

Abstract

Using a vision-inspired keyword spotting framework, we propose an architecture with input-dependent dynamic depth capable of processing streaming audio. Specifically, we extend a conformer encoder with trainable binary gates that allow us to dynamically skip network modules according to the input audio. Our approach improves detection and localization accuracy on continuous speech using Librispeech top-1000 most frequent words while maintaining a small memory footprint. The inclusion of gates also reduces the average amount of processing without affecting the overall performance. These benefits are shown to be even more pronounced using the Google speech commands dataset placed over background noise where up to 97% of the processing is skipped on non-speech inputs, therefore making our method particularly interesting for an always-on keyword spotter.