Relational Proxy Loss for Audio-Text based Keyword Spotting

TL;DR

The paper tackles audio-text based keyword spotting, where enrollment is text-driven and detection is audio-driven, by learning aligned acoustic and text embeddings. It introduces Relational Proxy Loss (RPL), which leverages the relational structure within each modality's embedding space (acoustic and text) rather than relying solely on point-to-point cross-modal comparisons. Experiments on the Wall Street Journal (WSJ) corpus show improved performance over standard deep metric learning losses such as triplet and proxy-based losses. The proposed approach provides a relational, structure-aware augmentation to cross-modal KWS with potential benefits for robustness and deployment in real-world keyword enrollment systems.

Abstract

In recent years, there has been an increasing focus on user convenience, leading to increased interest in text-based keyword enrollment systems for keyword spotting (KWS). Since the system utilizes text input during the enrollment phase and audio input during actual usage, we call this task audio-text based KWS. To enable this task, both acoustic and text encoders are typically trained using deep metric learning loss functions, such as triplet- and proxy-based losses. This study aims to improve existing methods by leveraging the structural relations within acoustic embeddings and within text embeddings. Unlike previous studies that only compare acoustic and text embeddings on a point-to-point basis, our approach focuses on the relational structures within the embedding space by introducing the concept of Relational Proxy Loss (RPL). By incorporating RPL, we demonstrated improved performance on the Wall Street Journal (WSJ) corpus.