Improving Rare Word Recognition with LM-aware MWER Training

TL;DR

This paper enhances rare word recognition in end-to-end speech models by integrating language models into discriminative training, leading to significant improvements in accuracy and more efficient inference.

Contribution

It introduces LM-aware MWER training for hybrid autoregressive transducer models, bridging the training-inference gap and improving rare word recognition performance.

Findings

10% relative improvement in voice search test sets with rare words

Achieves same WER in rescoring without sweeping fusion weights

Integrates LMs into discriminative training for better rare word recognition

Abstract

Language models (LMs) significantly improve the recognition accuracy of end-to-end (E2E) models on words rarely seen during training, when used in either the shallow fusion or the rescoring setups. In this work, we introduce LMs in the learning of hybrid autoregressive transducer (HAT) models in the discriminative training framework, to mitigate the training versus inference gap regarding the use of LMs. For the shallow fusion setup, we use LMs during both hypotheses generation and loss computation, and the LM-aware MWER-trained model achieves 10\% relative improvement over the model trained with standard MWER on voice search test sets containing rare words. For the rescoring setup, we learn a small neural module to generate per-token fusion weights in a data-dependent manner. This model achieves the same rescoring WER as regular MWER-trained model, but without the need for sweeping fusion weights.