Metric Learning for Keyword Spotting

TL;DR

This paper introduces a metric learning approach for keyword spotting that improves detection of target keywords while reducing false alarms from unseen non-target sounds, addressing limitations of traditional classifier-based methods.

Contribution

The work proposes a novel metric learning method with per-class weighting for keyword spotting, enhancing detection accuracy and reducing false alarms from unseen sounds.

Findings

Significantly reduces false alarms for unseen non-target keywords.

Maintains high classification accuracy on the Google Speech Commands dataset.

Outperforms traditional classifier-based keyword spotting methods.

Abstract

The goal of this work is to train effective representations for keyword spotting via metric learning. Most existing works address keyword spotting as a closed-set classification problem, where both target and non-target keywords are predefined. Therefore, prevailing classifier-based keyword spotting systems perform poorly on non-target sounds which are unseen during the training stage, causing high false alarm rates in real-world scenarios. In reality, keyword spotting is a detection problem where predefined target keywords are detected from a variety of unknown sounds. This shares many similarities to metric learning problems in that the unseen and unknown non-target sounds must be clearly differentiated from the target keywords. However, a key difference is that the target keywords are known and predefined. To this end, we propose a new method based on metric learning that maximises the distance between target and non-target keywords, but also learns per-class weights for target keywords \`a la classification objectives. Experiments on the Google Speech Commands dataset show that our method significantly reduces false alarms to unseen non-target keywords, while maintaining the overall classification accuracy.