Dou Ru. English Pronunciation Error Automatic Correction System Based on DTW Algorithm. Dynamic Systems and Applications 29 (2020) No. 5, 2058 – 2067

https://doi.org/10.46719/dsa202029540

ABSTRACT.
Our work aims to detect errors in pronunciation and provide feedback that can be found through speech processing and recognition methods. Automatic pronunciation error detection is used for two types of pronunciation errors, which are considered an error and language inheritance. This system integrates language knowledge and modern speech technology into the existing system. The training from Linguist Explanations can not only classify HMM to classify pros and cons sounds, but also convey how sounds are mispronounced. Phone errors are detected when arbitrary sounds and distorted rather than sound support the correct concept and precise pronunciation, high precision they produce with multimedia, and occur in L2 text. But there are some limitations in the technique of detecting its deviation. Therefore, it can improve the accuracy of distortion detection by studying the value of confidence as a result of HTK phonetic recognition. To be clear and precise inaccuracy, the less complex classic network, the proposed error methods have been focused on in limited terms, but the second unspecified online version of science fiction should be flexible enough to make mispronunciation. A 30-person recording corpus was tested using the dynamic time interruption (DTW) method in two classic science fiction timer types. It uses good pronunciation and mispronounced phonetic modeling through the Dynamic Interruption (DTW) algorithm. It involves changes to the word structure Further Section Search (FSS) is the Normalized Furthest Segment Search (NFSS). Different phonetic structures for each batch performance assessment ranging from a set of 12 words to 10 forms per training word, Polish correctly, and incorrectly pronounced words were conducted using sound info. The lowest average error rate (aerial) is DTW and NPDT (aerial = 0.287), whichever is better HMM has a better score than FSS (aerial = 0.473) combined. DTW quickly surpassed two modes of modeling HMMs. As far as a computer-assisted pronunciation training system is concerned, it is possible to work with a relatively small training corpus through speech training to mobilize technique support.

Keywords: Mispronunciation detection, dynamic time interruption algorithm. Automatic bug fix.