Text normalization based on deep learning considering similarities in terms of strings and sounds

Abstract

In this paper, we propose deep learning based models that can normalize text by considering the similarities of word strings and sounds. In the experiments, we compare the model that considers both the similarities of word strings and sounds, the model that considers only the similarity of word strings and of sounds, and the model without the similarities as a baseline model. As a result, all the proposed models achieved higher F1 score than the baseline model.