Sequence Discriminative Training for Deep Learning based Acoustic Keyword Spotting

TL;DR

This paper introduces a sequence discriminative training framework for deep learning based acoustic keyword spotting, significantly improving performance over previous frame-level methods in both fixed vocabulary and unrestricted tasks.

Contribution

It proposes novel sequence discriminative training approaches for acoustic KWS using word-independent lattices and non-keyword symbols, addressing a gap in existing research.

Findings

Achieved consistent performance improvements in fixed vocabulary KWS

Demonstrated significant gains in unrestricted KWS tasks

Validated effectiveness of sequence discriminative training over frame-level methods

Abstract

Speech recognition is a sequence prediction problem. Besides employing various deep learning approaches for framelevel classification, sequence-level discriminative training has been proved to be indispensable to achieve the state-of-the-art performance in large vocabulary continuous speech recognition (LVCSR). However, keyword spotting (KWS), as one of the most common speech recognition tasks, almost only benefits from frame-level deep learning due to the difficulty of getting competing sequence hypotheses. The few studies on sequence discriminative training for KWS are limited for fixed vocabulary or LVCSR based methods and have not been compared to the state-of-the-art deep learning based KWS approaches. In this paper, a sequence discriminative training framework is proposed for both fixed vocabulary and unrestricted acoustic KWS. Sequence discriminative training for both sequence-level generative and discriminative models are systematically investigated. By introducing word-independent phone lattices or non-keyword blank symbols to construct competing hypotheses, feasible and efficient sequence discriminative training approaches are proposed for acoustic KWS. Experiments showed that the proposed approaches obtained consistent and significant improvement in both fixed vocabulary and unrestricted KWS tasks, compared to previous frame-level deep learning based acoustic KWS methods.