Text Augmentation for Language Models in High Error Recognition Scenario

Karel Beneš, Lukáš Burget

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-wo…