A Chinese Text Correction and Intention Identification Method for Speech Interactive Context

摘要

ASR (Automatic Speech Recognition) is an important technology in man-machine interaction. Due to the complexity of natural language, the interference of environment and other factors, the recognition accuracy has low accuracy. This paper analyzes the use cases of speech recognition errors and proposes a text correction and intent recognition method based on the phonation principle and language characteristics peculiar to Chinese, and proposes an improved edit distance method to better calculate the text distance. Through a large number of experiments, this method can improves 22.9% accuracy of text recognition in ASR system.