Text normalization with convolutional neural networks

Abstract

Text normalization is a critical step in the variety of tasks involving speech and language technologies. It is one of the vital components of natural language processing, text-to-speech synthesis and automatic speech recognition. Convolutional neural networks (CNNs) have proven their superior performance to recurrent architectures in various application scenarios, like neural machine translation, however their ability in text normalization was not exploited yet. In this paper we investigate and propose a novel CNNs based text normalization method. Training, inference times, accuracy, precision, recall, and F1-score were evaluated on an open-source dataset. The performance of CNNs is evaluated and compared with a variety of different long short-term memory (LSTM) and Bi-LSTM architectures with the same dataset.