Research And Implementation Of Speech Intelligibility Evaluation Method Based On Phoneme

With the increasing frequency of global cultural exchanges,people are becoming more and more aware of the importance of mastering a foreign language.Oral training is the basic requirement for language learning,and computer-assisted pronunciation training system has also become an important research direction in speech recognition applications.An important indicator for evaluating the spoken English pronunciation of Chinese English learners is the intelligibility of speech,but there is a large gap between the relevance of automatic phonetic intelligibility scores at the phonetic level and the scores of human experts.In addition,due to the diversity of pronunciation errors of nonnative speaker learners,pronunciation of similar phonemes is prone to be confused during pronunciation detection of learners,and traditional detection methods based on probability and statistics are difficult to cover all types of erroneous phonemes.It is not possible to accurately provide phoneme-level correction suggestions to help improve speech intelligibility levels.In order to solve the above problems,this paper first analyzes the characteristics and limitations of the current common pronunciation detection techniques,in order to solve the problem of learners' incorrect pronunciation rules due to Chinese pronunciation habits.In this paper,a method based on phoneme biased pronunciation network is used to detect the phonemes of incorrect pronunciation in learner's pronunciation,and a circular random phoneme recognition detection method is proposed to improve the correctness and accuracy of phoneme recognition and use SVM classifier.Distinguish and classify the types of pronunciation errors,combined with the speech intelligibility evaluation methods,feedback feedback for the learner.Experiments show that using the improved phoneme recognition detection method in this paper improves the accuracy and accuracy of phoneme speech feature recognition.When assessing learner's speech intelligibility,in order to solve the problem that the commonly used speech intelligibility assessment method has a large gap with the relevance of human expert scores.This paper first studies the principle and characteristics of the evaluation method based on AI index,segment features and phoneme scores.A new phonetic intelligibility scoring method based on phoneme combination features is proposed,namely,a new acoustic feature measurement score is obtained through linear combination to score speech intelligibility,and the speech intelligibility scoring method is verified experimentally in this paper.Correlation with human expert ratings.At last,the paper applies the pronunciation problem detection method and speech intelligibility assessment method to the actual online prototype system,and designs a speech intelligibility assessment system for non-native speaker learners.Experimental tests show that after the system detects English pronunciation and corrects the incorrect phonemes,learners' speech intelligibility scores have been significantly improved,which further validates the validity and practicability of the proposed method.Therefore,the paper has certain reference value and application value in the field of computerassisted pronunciation.