Text Augmentation for Language Models in High Error Recognition Scenario

TL;DR

This paper explores data augmentation strategies for training speech recognition language models, finding that global error rate-based augmentation improves WER more effectively than other methods.

Contribution

It introduces a simple global error statistic-based augmentation scheme that outperforms label smoothing and improves speech recognition accuracy.

Findings

Global error-based augmentation improves WER from 1.1% to 1.9%.

Perplexity on augmented data does not predict final error rate.

Simple augmentation scheme outperforms more complex methods.

Abstract

We examine the effect of data augmentation for training of language models for speech recognition. We compare augmentation based on global error statistics with one based on per-word unigram statistics of ASR errors and observe that it is better to only pay attention the global substitution, deletion and insertion rates. This simple scheme also performs consistently better than label smoothing and its sampled variants. Additionally, we investigate into the behavior of perplexity estimated on augmented data, but conclude that it gives no better prediction of the final error rate. Our best augmentation scheme increases the absolute WER improvement from second-pass rescoring from 1.1 % to 1.9 % absolute on the CHiMe-6 challenge.