Joint Audio/Text Training for Transformer Rescorer of Streaming Speech Recognition

TL;DR

This paper introduces a joint audio/text training approach for Transformer Rescorers in streaming speech recognition, enabling the use of unpaired text data to improve accuracy without increasing latency or model complexity.

Contribution

The novel joint training method leverages unpaired text data for Transformer Rescorers, reducing data costs and enhancing recognition accuracy in streaming ASR systems.

Findings

Significant WER reduction on Librispeech and in-house datasets.

Effective use of unpaired text data without extra latency.

Improved rescoring performance over standard Transformer Rescorer.

Abstract

Recently, there has been an increasing interest in two-pass streaming end-to-end speech recognition (ASR) that incorporates a 2nd-pass rescoring model on top of the conventional 1st-pass streaming ASR model to improve recognition accuracy while keeping latency low. One of the latest 2nd-pass rescoring model, Transformer Rescorer, takes the n-best initial outputs and audio embeddings from the 1st-pass model, and then choose the best output by re-scoring the n-best initial outputs. However, training this Transformer Rescorer requires expensive paired audio-text training data because the model uses audio embeddings as input. In this work, we present our Joint Audio/Text training method for Transformer Rescorer, to leverage unpaired text-only data which is relatively cheaper than paired audio-text data. We evaluate Transformer Rescorer with our Joint Audio/Text training on Librispeech dataset as well as our large-scale in-house dataset and show that our training method can improve word error rate (WER) significantly compared to standard Transformer Rescorer without requiring any extra model parameters or latency.