Design of an Automatic English Pronunciation Error Correction System Based on Radio Magnetic Pronunciation Recording Devices

Abstract

In this paper, a system for automatic detection and correction of mispronunciation of native Chinese learners of English by speech recognition technology is designed with the help of radiomagnetic pronunciation recording devices and computer-aided software. This paper extends the standard pronunciation dictionary by predicting the phoneme confusion rules in the language learner’s pronunciation that may lead to mispronunciation and generates an extended pronunciation dictionary containing the standard pronunciation of each word and the possible mispronunciation variations, and automatic speech recognition uses the extended pronunciation dictionary to detect and diagnose the learner’s mispronunciation of phonemes and provides real-time feedback. It is generated by systematic crosslinguistic phonological comparative analysis of the differences in phoneme pronunciation with each other, and a data-driven approach is used to do automatic phoneme recognition of learner speech and analyze the mapping relationship between the resulting mispronunciation and the corresponding standard pronunciation to automatically generate additional phoneme confusion rules. In this paper, we investigate various aspects of several issues related to the automatic correction of English pronunciation errors based on radiomagnetic pronunciation recording devices; design the general block diagram of the system, etc.; and discuss some key techniques and issues, including endpoint detection, feature extraction, and the system’s study of pronunciation standard algorithms, analyzing their respective characteristics. Finally, we design and implement a model of an automatic English pronunciation error correction system based on a radiomagnetic pronunciation recording device. Based on the characteristics of English pronunciation, the correction algorithm implemented in this system uses the similarity and pronunciation duration ratings based on the log posterior probability, which combines the scores of both, and standardizes this system scoring through linear mapping. This system can achieve the purpose of automatic recognition of English mispronunciation correction and, at the same time, improve the user’s spoken English pronunciation to a certain extent.

AI and AI-powered tools for pronunciation training

“…AI systems are based on the system's ability to imitate human behaviour (Almelhes, 2023) and even learn from interactions with humans and other systems. In language education, AI is based on natural language processing, speech synthesis and recognition (Shufang, 2021). In practice, AI tools analyse human linguistic behaviour, create their own language model and apply it in communication with human speakers.…”

“…The origins of using technology to calculate deviations from a model pronunciation trace back to the 1990s (Shufang, 2021) when the then-used system evaluated features of segments and sentences that were not prespecified. However, in the early 2000s, the systems based on ASR evaluated pronunciation differently from human raters across more prolonged stretches of spoken texts, while the evaluation of single phonemes was similar.…”

“…However, although much research has been done into the system's functionality, relatively little in comparison has been done to investigate pedagogical opportunities and challenges it brings in pronunciation training. Shufang (2021) refers to this as human-computer interaction (or human-machine communication), which comprises elements of speech recognition, natural language understanding, and speech synthesis. Speech synthesis, in particular, is the bridging element between this type of communication and artificial intelligence.…”

“…Thus, pronunciation training and the subsequent evaluation through AI-powered tools are most frequently carried out using commercially available learning applications and programs. Non-specified systems are being tested for accuracy by a non-defined group of speakers (Jiao et al, 2021;Shufang, 2021;Cheng & Wang, 2022) and indicate future improvements of the tools.…”

Application of Recurrent Neural Networks Based on Attentional Mechanisms in Classification Error Correction in English Teaching

“…In recent years, automatic error correction with the help of automated algorithms has gradually become a trend in English teaching. Literature [12] proposed an extension method for the prediction of phoneme confusion rules based on mispronunciation for the standard pronunciation lexicon and constructed an on-mapping method between mispronunciation and standard pronunciation to realize the error detection and correction of English pronunciation. Literature [13] is oriented to the oral expression of English, applies the model of communicative language proficiency to multimodal oral teaching data, constructs a matrix model of English oral communicative proficiency, and is able to correct the oral expression for the user based on preferences.…”