Automatic Spelling Correction for ASR Corpus in Traditional Chinese Language using Seq2Seq Models

The goal of Automatic Speech Recognition (ASR) service is to translate spoken language into text. There exist many factors that will degrade the performance of the ASR system, such as environmental noise, human pronunciation, etc. This research focuses on automatic spelling correction for traditional Chinese corpus generated by ASR systems. We show that a Sequence to Sequence (Seq2Seq) neural network model with attention mechanism could be improved by adding pointer network with auxiliary phoneme features of the input word sequence.