Research on Text Error Correction Algorithm after Automatic Speech Recognition Based on Pragmatic Information

Abstract

Error correction for automatic speech recognition text is an indispensable part of artificial intelligence. At present, speech to text (STT) has widely requirements for the processing of pragmatic information. The text correct rate in STT is the foundation for NLP. Aiming at the text error problems of traditional error correction methods that cannot understand semantics and sentence meanings well. The proposed method used the long and short-term memory neural network (LSTM) algorithm with monte-carlo tree search in this paper. The text error leads to mistake in semantic slot filling for NLP. Therefore, the proposed combined algorithm and optimization method solved the problem by experiments. The results verified the accuracy increased 25% for the telephone inquiry by text error correction.