Improved Neural Language Model Fusion for Streaming Recurrent Neural Network Transducer

TL;DR

This paper introduces an enhanced fusion technique for RNN-T models that effectively incorporates external neural network language models during training and inference, leading to significant WER improvements without added latency.

Contribution

It extends existing fusion methods to enable RNN-T to utilize external NNLMs during both training and inference, improving accuracy while maintaining low latency and flexibility.

Findings

13-18% relative WER reduction on Librispeech

No additional latency introduced

Flexible plug-and-play of NNLMs without re-training

Abstract

Recurrent Neural Network Transducer (RNN-T), like most end-to-end speech recognition model architectures, has an implicit neural network language model (NNLM) and cannot easily leverage unpaired text data during training. Previous work has proposed various fusion methods to incorporate external NNLMs into end-to-end ASR to address this weakness. In this paper, we propose extensions to these techniques that allow RNN-T to exploit external NNLMs during both training and inference time, resulting in 13-18% relative Word Error Rate improvement on Librispeech compared to strong baselines. Furthermore, our methods do not incur extra algorithmic latency and allow for flexible plug-and-play of different NNLMs without re-training. We also share in-depth analysis to better understand the benefits of the different NNLM fusion methods. Our work provides a reliable technique for leveraging unpaired text data to significantly improve RNN-T while keeping the system streamable, flexible, and lightweight.