Research on Intelligent Enhancement Methods for Speech Recognition Text Data

This paper addresses the need to improve the readability of speech recognition transcribed text by designing the speech recognition readability enhancement solution. It can better process the transcribed text from speech recognition and generate high-readability text for downstream applications. The paper proposes a method for synthesizing a readability dataset for speech recognition, and implements a new data augmentation method to solve the problem of the lack of dedicated training data for speech recognition readability research. In addition, a stage-based new training method is designed to better utilize the training data to improve model performance. Experimental results show that the speech recognition readability enhancement solution designed in this paper is able to generate text with higher readability than the related studies for speech recognition transcription.