手机语音识别技术对普通话儿童声母发音客观评估的初步研究  

A preliminary study: objective assessment of initial constant pronunciation of Mandarin-speaking children based on mobile phone speech recognition

作　　者：韩源[1] 姜成惠[1] 李盛[1] 汪彬昺[1] 王雨晨[1] 邱秋[1] 施星辉[1] 

机构地区：[1]南京医科大学口腔疾病研究江苏省重点实验室南京医科大学附属口腔医院口腔颌面外科,江苏南京210029

出　　处：《口腔医学》2017年第6期515-519,共5页Stomatology

基　　金：江苏省科技厅省自然科学基金(BK2009347);江苏省高校优势学科建设工程(2014-037)

摘　　要：目的通过探讨手机语音识别技术对构音正常儿童声母发音的识别率,研究目前较为常见的手机语音识别软件对声母识别效果,为探索语音识别软件在构音障碍儿童语音评估和矫正中的应用提供基础。方法应用基于手机i OS系统的两种语音识别软件(XF~@和WX~@)对31例构音正常的儿童进行语音评估,计算识别率,运用SPSS 19.0进行配对t检验,比较不同语音识别软件的识别率。将目标音根据发音方式与发音部位分类,运用多个率卡方检验与Scheffe可信区间法筛识别率较低的发音类型。结果 WX~@语音识别率为(86.84±3.90)%、XF~@语音识别率(89.01±4.00)%,经统计学处理发现两种识别软件的识别率存在显著差异(P<0.05)。根据发音方式分类,不同声母的识别率存在显著差异,送气塞擦音识别率(82.88%)低于其他部分发音。根据发音部位分类,不同声母的识别率亦存在差异,舌尖前音(92.56%)与舌尖后音(95.15%)低于其他部分发音。结论不同手机语音识别软件与发音类型对构音正常儿童声母发音的识别率有影响。本研究结果对语音识别技术在语音评估及治疗领域的应用提供一定依据与参考。Objective To evaluate the validity of mobile phone speech recognition software on initial consonant production by typical Mandarin-speaking children, and to explore the application of mobile-phone-based speech recognition software on speech assessment and intervention. Methods Speech recognition rate of initial consonants produced by 31 children was calculated using two kinds of speech recognition software （XF^@ and WX^@ ）. A paired t test was performed to compare the difference between the two systems using SPSS19.0.A multiple rate chi-square test and a Scheffe＇s confidence interval test were used to screen the articulation type with low speech recognition rate according to different places and manners of articulation. Results Speech recognition rate of XF^@ （89.01 ± 4.00）%was significantly higher than that of WX^@（ 86.84± 3.90）% and there was statistically significant difference （P〈 0.05 ）. According to the manner of consonant, speech recognition rate of aspirated affricate was significantly lower （82.88%） than other manners of articulation. According to the place of consonant, speech recognition rate of the dental （ 92.56% ） and alveolar （ 95.15% ） were significantly lower compared to other places of articulation. Conslusion Type of software and classification of consonants affect the speech recognition rate of initial consonant production of children who can normally construct the sound. This study provides some preliminary data for the application of mobile-phone-based speech recognition software, which merits further investigation.

关 键 词：语音识别 普通话儿童 语音评估 语音治疗 声母 

分 类 号：H017[语言文字—语言学]