本硏究係探討聽覺障礙學生對不同說話者所說出的韻母的讀話辨識情形，及讀話所 見的視素有無差異。受試爲卅名就讀台北市國民小學啓聰班及資源教室的聽覺障礙學生。 本硏究所用的硏究工具爲自編之「注音符號讀話能力評量表韻母篇」，以十四個韻母 (廿儿兩音除外）和兩個聲母（厂了）排列組合而成。所得資料以群落分析和單因子變 異數分析進行處理。本硏究的主要發現如下：1受試只能將十四個韻母區分出一個視素： 共包含九個韻母〔亡巧'幺只尤厶〕。其他五個韻母〔丫艺一Xu〕未能凝聚成視 素。2.不同說話者形成不同的視素。觀看乙說話者所形成的視素中有十個韻母（包括丫 音）。3.視素中韻母凝聚的原因與其發音特性有關。4.韻母之平均讀話正確率平均爲 47.54%，各個韻母讀話正確率從77.50%到15.00%不等。觀看幺只一X四音的正確率最 高，都在六成以上，但許多音的辨識率都未達半數，特別是亡的正確率只達15 %。5. 韻母視素讀話正確率爲84.03%.。同一視素内的韻母易有視覺混淆的特性。不同說話 者導致的讀話正確率有所不同。整體而言甲說話者最易於辨識，乙、丙說話者則無太大 差異。但只有汚'么巧卜太_U等音的得分有顯著差異。

The purposes of this study were to explore the contents of vowel visemes lipread by elementary . students with hearing impairment, and to examine whether talker differences existed by way of different visemes revealed. Subjects were thirty elementary students from four resource programs for the hearing-impaired in Taipei City. They were all 6th graders. Nineteen of them were boys and eleven were girls. They were pre-lingual deaf with severe to profound hearing loss. Twenty-two of them had normal eyesight, while the other eight wore glasses. T^hey lipread a set of 156 CV syllables and then wrote down what they thought they had seen. The syllables were formed by one consonant (either /n/ or /h/) plus one of the 14 Manderine vowels (with the exclusion of two vowels for articulatory reasons). These syllables were divided into three parts, and each part was spoken by one speaker. They were videotaped for the purpose of this study. Data were analyzed with Agglomerative hierarchical cluster analysis and ANOVA. The results were as follows: 1. Only one cluster was formed, with nine vowels involved. There were five vowels which failed to form a viseme. 2. Percentage of correct response for that cluster was high: 84.03%, while that for each consonant varied from 77.50% to 15.00%, with the average 47.54%. It also showed that most of the errors were located within the very cluster, which indicated that vowels within the cluster were highly confused. 3. There seemed to be no significant differences in viseme formed across three talkers; minute discrepancies, however, did emerge in two areas: (1) ten vowels were clustered by lipreading B speaker; and (2) scores of eight vowels showed significant differences.