Spoken English speech recognition and evaluation method and system, computer and readable storage medium

Abstract

The invention provides a spoken English speech recognition and evaluation method, a computer and a readable storage medium, and relates to the field of pattern recognition and natural language understanding, the spoken English speech is preprocessed, and the noise of the audio is suppressed by using an LMS adaptive filter noise reduction method in the preprocessing process to obtain a preprocessed audio sequence; predicting the audio sequence to obtain probability distribution of predicted characters; decoding the obtained probability distribution of the predicted characters to obtain a speech recognition result; inputting the obtained probability distribution of the predicted characters into a GOP-based spoken English speech evaluation method for evaluation to obtain a pronunciation score result; and introducing the voice data scored by the expert, and carrying out comprehensive scoring on the voice. Compared with the prior art, the terminal-to-terminal speech recognition is adopted, the recognition accuracy is obviously improved, the early-stage work of audio-to-text and speech evaluation can be realized, the calculation efficiency is high, and the evaluation accuracy is ensured.

Description

technical field [0001] The present invention relates to the field of pattern recognition and natural language understanding, in particular to a deep learning-based spoken English speech recognition and evaluation method, system, computer and readable storage medium. Background technique [0002] In recent years, with the rapid development of pattern recognition and natural language understanding, and the in-depth application of machine learning, especially deep learning and other technologies, the research and application fields of speech recognition and speech evaluation technology have become more and more extensive. Speech recognition technology transcribes speech signals into corresponding text or commands through computers, which is essentially a process of pattern recognition. Speech evaluation technology is a technology that automatically evaluates, scores, detects errors, and provides corrections for the oral pronunciation level and pronunciation. The evaluation dime...

AI Technical Summary

Problems solved by technology

Among them, RNN is widely used in acoustic models because it can capture the sequential dependence information of sequence data, but the calculation of RNN at each moment requires the output of the previous moment as input, so it can only be calculated serially, and the speed is very slow , in addition, RNN is vulnerable to gradient disappearance during training, converges more slowly, and requires more computing resources

[0004] In the prior art, there is also a waveform comparison between the human voice and the standard original sound. Although the real-time rate can be guaranteed, the accuracy is low, and all words must have a corresponding standard original sound, which is very inflexible.

Images