Boosting Tail Neural Network for Realtime Custom Keyword Spotting

TL;DR

This paper introduces a Boosting Tail Neural Network (BTNN) to enhance real-time custom keyword spotting, achieving better accuracy and lower false alarms with limited computational resources, inspired by brain activation principles.

Contribution

The paper presents a novel BTNN architecture that leverages boosting of weak classifiers to improve keyword spotting performance in real-time applications.

Findings

18% relative improvement over traditional single-classifier methods

Better wakeup rate and reduced false alarms

Potential for future automatic speech recognition applications

Abstract

In this paper, we propose a Boosting Tail Neural Network (BTNN) for improving the performance of Realtime Custom Keyword Spotting (RCKS) that is still an industrial challenge for demanding powerful classification ability with limited computation resources. Inspired by Brain Science that a brain is only partly activated for a nerve simulation and numerous machine learning algorithms are developed to use a batch of weak classifiers to resolve arduous problems, which are often proved to be effective. We show that this method is helpful to the RCKS problem. The proposed approach achieve better performances in terms of wakeup rate and false alarm. In our experiments compared with those traditional algorithms that use only one strong classifier, it gets 18\% relative improvement. We also point out that this approach may be promising in future ASR exploration.