Complementary Language Model and Parallel Bi-LRNN for False Trigger Mitigation

TL;DR

This paper introduces a novel false trigger mitigation approach for voice assistants using a parallel decoding process with a specialized out-of-domain language model and a combined Bi-LRNN classifier, significantly reducing false triggers.

Contribution

It proposes a parallel ASR decoding with a specialized language model and a combined Bi-LRNN classifier, achieving substantial false trigger rate reduction.

Findings

38.34% reduction in false trigger rate at 0.4% false suppression

Further 10.8% reduction with combined lattice-based Bi-LRNN

Effective mitigation of false triggers in voice assistants

Abstract

False triggers in voice assistants are unintended invocations of the assistant, which not only degrade the user experience but may also compromise privacy. False trigger mitigation (FTM) is a process to detect the false trigger events and respond appropriately to the user. In this paper, we propose a novel solution to the FTM problem by introducing a parallel ASR decoding process with a special language model trained from "out-of-domain" data sources. Such language model is complementary to the existing language model optimized for the assistant task. A bidirectional lattice RNN (Bi-LRNN) classifier trained from the lattices generated by the complementary language model shows a $38.34\%$ relative reduction of the false trigger (FT) rate at the fixed rate of $0.4\%$ false suppression (FS) of correct invocations, compared to the current Bi-LRNN model. In addition, we propose to train a parallel Bi-LRNN model based on the decoding lattices from both language models, and examine various ways of implementation. The resulting model leads to further reduction in the false trigger rate by $10.8\%$.