Open AccessSpoken English Repetitive Correction Retrieval Based on Syllable Unit WFST Web Search Filter
Author(s) -
Shengquan Yu
Publication year - 2020
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1533/3/032010
Subject(s) - computer science , pattern recognition (psychology) , transformer , filter (signal processing) , search algorithm , algorithm , artificial intelligence , speech recognition , computer vision , physics , quantum mechanics , voltage
Aiming at the difficulty in computer-assisted language teaching of repeated correction and error detection in spoken English, a fault-tolerant alignment and search filtering algorithm based on the syllable unit WFST (weighted finitestate transformer) network is proposed. This algorithm uses the script corresponding to the adjacent matching words in the recognition results to establish the above grammatical network for fault-tolerant alignment under secondary recognition. The candidate modified parts and replacement parts are used as the query and template for search filtering. In the end, the result of repeated correction error detection is determined by the confidence of the search filtering algorithm. To this end, the k-difference algorithm based on sequential assumptions and the n-gram algorithm based on random assumptions are proposed. Experiments show that the multi-n-gram hybrid search filtering with syllables as the modeling unit achieves relatively optimal results without using quadratic fault-tolerant alignment; when using quadratic fault-tolerant alignment, F-measure can obtain 3 to 4 Further increase in percentage.