Research on the Construction of English Translation Model for Speech Recognition Based on Multiple Information Sources and Lexical Assistance

Abstract

Speech recognition and machine translation technologies have been continuously evolving and improving to meet the growing demand for multilingual communication. Meanwhile, diverse information source inputs make traditional speech recognition and translation systems even more challenging. In view of this, the study takes multiple information sources as an entry point to improve the existing English speech recognition and translation systems. Aiming at the process of speech comprehension and text generation in speech recognition, the study utilizes the lexical labels in the lexical auxiliary function to make functional adjustments to the input of the speech source and the output of the translated text. Secondly, after incorporating multiple information source conditions, the study proposes a novel English speech recognition translation model. The experimental results show that compared with other speech recognition models, the new model proposed by the study recognizes up to 87% average accuracy. In addition, the research proposed model can further improve the BLEU scores of the translations and strengthen the textual information with a maximum score of 0.81. In summary, lexical assistance under multiple information sources can significantly improve the performance of English translation models for speech recognition, providing new ideas and methods for the development of speech recognition and translation technology.