An Automatic Diagnosis Method for Chinese Speech Pronunciation Errors Based on Deep Full-Sequence Convolutional Neural Network
Artificial intelligence algorithms have demonstrated potential in the intelligent educational ecosystem of teaching Chinese as a second language, but their application in the field of automatic diagnosis of speech pronunciation errors is often limited by insufficient mining of pronunciation features of non-native speakers and poor diagnostic reliability. To overcome this dilemma, an automatic diagnosis method of Chinese pronunciation errors for non-native speakers is proposed. By extracting the pronunciation features of non-native speakers, judging the correctness of pronunciation based on pronunciation goodness scores and setting thresholds, and using deep full-sequence convolutional neural network modeling to optimize parameters, the automatic diagnosis of Chinese pronunciation errors for non-native speakers is achieved. Experiments show that the recall rate of normal pronunciation diagnosis in the confusion matrix of the research method reaches 94.7% and the precision rate is 95.1%. In terms of the coverage of error types, the coverage rate of level tone misreading in tone errors reaches 87.6%, the coverage rate of aspirated and unaspirated initial errors reaches 90.3%, and the coverage rate of single vowel pronunciation errors in final errors reaches 89.1%. Experiments show that the research method significantly improves the accuracy and comprehensiveness of Chinese pronunciation error diagnosis by deeply mining and accurately modeling the pronunciation characteristics of non-native speakers.