Multiple-Instance, Cascaded Classification for Keyword Spotting in Narrow-Band Audio

TL;DR

This paper introduces a cascaded deep neural network approach for keyword spotting in narrow-band 8kHz audio, effectively handling class imbalance and reducing power consumption in non-IID environments.

Contribution

The work presents a novel cascaded classifier system combining multiple features and multiple-instance learning for improved keyword spotting in challenging audio conditions.

Findings

False negative rate of 6% achieved

False positive rate of 0.75 per hour

Reduced power consumption via early termination

Abstract

We propose using cascaded classifiers for a keyword spotting (KWS) task on narrow-band (NB), 8kHz audio acquired in non-IID environments -- a more challenging task than most state-of-the-art KWS systems face. We present a model that incorporates Deep Neural Networks (DNNs), cascading, multiple-feature representations, and multiple-instance learning. The cascaded classifiers handle the task's class imbalance and reduce power consumption on computationally-constrained devices via early termination. The KWS system achieves a false negative rate of 6% at an hourly false positive rate of 0.75