基于隐马尔科夫模型的维吾尔语腭裂患者术后语音理解度自动评估研究

发布时间：2018-01-11 22:05

本文关键词：基于隐马尔科夫模型的维吾尔语腭裂患者术后语音理解度自动评估研究　出处：《新疆医科大学》2017年硕士论文　论文类型：学位论文

【摘要】：目的:为维吾尔语腭裂患者术后语音理解度的评估,提供一种客观的辅助判别手段。方法:从维吾尔族儿童跟读语音语料库中选取150名正常儿童语音,其中随机选择120例作为训练数据用于识别模型的建立,选择30例作为测试数据(正常组)。从维吾尔族腭裂术后患者跟读语音语料库中选取30名腭裂术后腭咽闭合不全患儿的语音资料(患者组)。提取训练数据的语音特征参数,使用隐马尔科夫模型工具包建立语音识别模型,将此模型对腭裂患者术后语音理解度的自动评估结果与3名专家的主观评估结果进行比较。结果:基于隐马尔科夫模型的语音识别模型对正常语音的正确识别率较高(91.28±3.60%),对腭裂术后语音正确识别率较低(35.94±11.67%),差异具有统计学意义(P0.05)。自动评估结果与作为金标准的主观评估对腭裂患者术后语音理解度的评估结果之间相关系数为0.848,具有高度的相关性。结论:基于隐马尔科夫模型的维吾尔语腭裂患者术后语音理解度自动评估可以作为主观评估的辅助手段,对腭咽闭合功能的判断提供一定的参考。
[Abstract]:Objective: to evaluate the speech comprehension of Uygur patients with cleft palate. Methods: 150 normal children were selected from Uygur children's phonetic corpus. Among them, 120 cases were randomly selected as training data to establish the recognition model. 30 patients were selected as test data (normal group). 30 children with palatopharynx incompetence after cleft palate operation were selected from Uygur cleft palate patients' following speech corpus (patient group). Extract the speech feature parameters of the training data. Using Hidden Markov Model Toolkit to build speech recognition model. The results of automatic assessment of postoperative speech comprehension of patients with cleft palate were compared with the subjective results of 3 experts. The correct recognition rate of speech recognition model based on Hidden Markov Model is higher than that of normal speech (. 91.28 卤3.60). The rate of correct speech recognition after cleft palate surgery was 35.94 卤11.67). The correlation coefficient between the results of automatic assessment and subjective evaluation as gold standard for postoperative speech comprehension of patients with cleft palate was 0.848. Conclusion: the automatic evaluation of speech comprehension of Uygur cleft palate patients based on Hidden Markov Model can be used as a supplementary method for subjective evaluation. To the palatopharynx closure function judgment provides the certain reference.
【学位授予单位】：新疆医科大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R782.2

【参考文献】