External Language Model Integration for Factorized Neural Transducers

TL;DR

This paper introduces an adaptation method for factorized neural transducers that effectively integrates external language models, significantly improving speech recognition accuracy through linear interpolation and class-based n-gram models.

Contribution

It presents a novel approach for integrating external language models into FNT, demonstrating substantial accuracy gains over previous methods.

Findings

Linear interpolation of external LMs outperforms shallow fusion.

Class-based n-gram models improve FNT accuracy.

Up to 60% WERR gain in entity-rich scenarios.

Abstract

We propose an adaptation method for factorized neural transducers (FNT) with external language models. We demonstrate that both neural and n-gram external LMs add significantly more value when linearly interpolated with predictor output compared to shallow fusion, thus confirming that FNT forces the predictor to act like regular language models. Further, we propose a method to integrate class-based n-gram language models into FNT framework resulting in accuracy gains similar to a hybrid setup. We show average gains of 18% WERR with lexical adaptation across various scenarios and additive gains of up to 60% WERR in one entity-rich scenario through a combination of class-based n-gram and neural LMs.