A Study on the Translation of Spoken English from Speech to Text

Abstract

Rapid translation of spoken English is conducive to international communication. This paper briefly introduces a convolutional neural network (CNN) algorithm for converting English speech to text and a long short-term memory (LSTM) algorithm for machine translation of English text. The two algorithms were combined for spoken English translation. Then, simulation experiments were performed by comparing the speech recognition performance among the CNN algorithm, the hidden Markov model, and the back-propagation neural network algorithm and comparing the machine translation performance with the LSTM algorithm and the recurrent neural network algorithm. Moreover, the performance of the spoken English translation algorithms combining different recognition algorithms was compared. The results showed that the CNN speech recognition algorithm, the LSTM machine translation algorithm and the combined spoken English translation algorithm had the best performance and sufficient anti-noise ability. In conclusion, utilizing a CNN for converting English speech to texts and LSTM for machine translation of the converted English text can effectively enhance the performance of translating spoken English.