基于模糊综合评价的语音评测模型的研究与实现

发布时间：2018-01-15 13:04

本文关键词：基于模糊综合评价的语音评测模型的研究与实现　出处：《东华理工大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着移动互联网的迅速发展,语音评测技术已经成为当前国内外口语学习领域的研究热点,目前主要应用于教育行业中的口语考试评测和口语辅助教学当中。在外语学习过程中,口语学习一直是众多外语学习者的软肋,因此如何有效的利用语音评测技术解决外语口语不达标的问题成为当前研究的重点和难点。本文针对目前英语学习中口语评测问题,在分析了英语发音特点以及语音韵律特征的基础上,提出了基于模糊综合评价的语音评测模型,同时将该方法进行移植并实现了基于Android平台的语音评测系统。主要工作内容如下:首先建立基于云平台下的标准语音评测库。采用英国和美国国家语料库作为标准语音库。通过线上收集的方法抓取网页上英语语音文件并分类保存到云平台上,以及通过线下收集方法对英语单词进行分类后收集了10个人(不同性别)读300个单元音的4种情况(正常、升调、降调、错读)、300个双元音、600个多元音的7种情况(正常、快读、慢读、升调、降调、部分错、全部错)总共75000个语音文件数据作为语音训练库,由10位英语专家根据发音情况进行评测并建立专家评测库,通过Android平台的语音上传模块以及语音评测模块不断扩充云平台的语音测试库。其次在分析英语发音中音高、音长和轻重音等发音特点的基础上提取语音的韵律特征信息(梅尔频率倒谱系数特征、基频特征和短时能量特征),利用层次分析法计算出各特征参数对语音评测结果的权重值,采用动态时间规则算法对标准语音和测试语音进行相似度比对,同时提出了基于模糊综合评价的语音评测模型并给出最后语音评测结果。通过Matlab仿真实验对改进的语音评测模型进行详细的性能分析以及实验对比,实验显示该改进模型有很强的高效性和实用性。最后将改进的语音评测模型移植到Android平台上,设计并实现了基于云平台的语音评测系统,通过实验人员对系统进行评测效果测试,测试结果基本可以满足语音评测系统要求。
[Abstract]:With the rapid development of mobile Internet, voice evaluation technology has become a hot topic in the field of oral English learning at home and abroad. At present, it is mainly used in the evaluation of oral examination and oral assistant teaching in education industry. In the process of foreign language learning, oral learning has always been the weakness of many foreign language learners. Therefore, how to effectively use the phonetic evaluation technology to solve the problem of oral English substandard has become the focus and difficulty of current research. This paper aims at the problem of oral evaluation in English learning. Based on the analysis of English pronunciation and phonetic prosody, a speech evaluation model based on fuzzy comprehensive evaluation is proposed. At the same time, the method is transplanted and the voice evaluation system based on Android platform is realized. The main work is as follows:. Firstly, the standard voice evaluation database based on cloud platform is established. The national corpus of the United Kingdom and the United States is used as the standard voice database. The English voice files on the web page are captured and stored on the cloud platform by the method of online collection. After classifying English words offline, 10 people (of different genders) were collected in 4 situations (normal, ascending, descending, misreading, 300 divowels). A total of 75000 voice file data as a voice training library for 7 cases of 600 polyvowels (normal, fast, slow, ascending, down, partially wrong, all wrong). According to the pronunciation of 10 English experts to evaluate and establish an expert evaluation database. Through the voice upload module of Android platform and the voice evaluation module, the voice test library of cloud platform is expanded. Secondly, the analysis of the high pitch of English pronunciation is made. The prosodic feature information (Mel frequency cepstrum coefficient feature, fundamental frequency feature and short time energy feature) is extracted based on the phonetic characteristics such as length and light and heavy tone. AHP is used to calculate the weight of each characteristic parameter to the result of speech evaluation, and the dynamic time rule algorithm is used to compare the similarity between the standard speech and the tested speech. At the same time, the speech evaluation model based on fuzzy comprehensive evaluation is proposed and the final speech evaluation results are given. The improved speech evaluation model is analyzed and compared in detail by Matlab simulation experiment. Experiments show that the improved model has high efficiency and practicability. Finally, the improved voice evaluation model is transplanted to Android platform, and a voice evaluation system based on cloud platform is designed and implemented. The test results can meet the requirements of the speech evaluation system.
【学位授予单位】：东华理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN912.3

【相似文献】