華語捲舌音對比的聽覺辨識與頻譜動差分析

The Auditory Discrimination of Mandarin Retroflex Contrasts and Spectral Moment Analysis

華語中有三對摩擦音和塞擦音的捲舌／非捲舌音對比，華語捲舌音具有特殊的聲學和聽知覺特徵。本研究的目的主要在探討正常成年人對於單音節和雙音節詞中捲舌音對比的聽覺區辨以及聲學頻譜動差參數和捲舌音與非捲舌音聽辨之間的關係。共有三十個正常成人參與捲舌音判斷的實驗，實驗包含兩個作業，第一個作業使用具有三種捲舌程度的單音節詞為聽覺刺激，包括不捲舌、稍捲舌和較捲舌三類，要求聽者做是否為捲舌音的判斷。第二個作業刺激材料為雙音節詞，聽者根據第一音節的聽知覺捲舌性做判斷。單音節方面的結果發現，若稍捲舌音不計，聽者的總正確率為78%。對三種捲舌程度刺激的反應時間比較，對最捲舌刺激的反應時間顯著最短；而稍捲舌和不捲舌之間則無顯著差異。根據聽者的反應定義刺激的捲舌對比類別，結果顯示捲舌音類比起非捲舌音類刺激有顯著較低的Ml，較高的M2，較高的M3，和較低的M4，這些趨勢和之前的捲舌音製造聲學分析研究（鄭靜宜，2006）的資料相吻合。四個頻譜動差參數中捲舌音反應百分比和M1的相關值最高，達-.61 （p < .001）。在作業二中使用雙音節刺激來檢視詞彙脈絡效果，結果發現聽者的捲舌音判斷既非完全是根據詞彙線索，也非單就聲學線索，而是兩者因素的綜合，而聲學線索中第一個頻譜動差參數（M1）在區分華語雙音節詞捲舌音對比仍為有效的參數。

There are three pairs of retroflex contrasts in Mandarin fricative and affricates, and perceptually, they are very similar and not easy to be distinguished, especially for non-skilled listeners. The purpose of this study is to investigate the auditory perception of retroflexation of Mandarin in adults and investigate the relationship between the spectral moments of frication noise and the auditory discrimination of the fricatives and affricates with retroflexed contrasts. There were thirty adults of native Mandarin speakers participating in the retroflex recognition tasks. There were two tasks of retroflexation judgment in the experiment. In the first one task, the monosyllabic words were used as stimuli and the listeners were asked to judge the retroflexation of Mandarin monosyllabic words. In the second task, the disyllabic words were used, and the retroflexation of the first syllables of words was required to be judged. In the first task, the monosyllable stimuli had three degrees of retroflexation variations which was consciously made by the speakers, including nonretroflexation, slightly retroflexation, and more retroflexation. The results show that if the slightly retroflexed stimuli were excluded, the correctness percentage of responds was 70% for all listeners. The highly retroflexed consonants had the significantly shortest reaction time (RT), while the RTs between the nonretroflexed and the slightly retroflexed had no significant difference. According to the retroflexness categorical responses of the listeners, the spectral moment analysis of the stimuli showed that the retroflexed consonants had significantly lower M1, higher M2, higher M3, and lower M4, compared to the nonretroflxed consonants. The results are compatible with the data of the previous study of retroflex production (Jeng, 2005). Among the four spectral moments, M1 was the most robust parameter for discriminating the retroflex contrasts The correlation coefficient between the retroflex response percentages and M1 of the stimuli was -0.61 (p < .0001). In the second task, the listeners were asked to judge the retroflexation of the first syllables of the disyllabic words based on their acoustical retroflexation, instead of their lexical retroflexation. However the results show that the retroflexation judgment of the listeners was inevitably influenced by the lexical information of the second syllables of the words. Among the four spectral moments, M1 still the most robust parameter for discriminating the retroflex contrasts. The interaction of the lexical information and acoustical information during the retroflexation judgment of the disyllabic words was discussed.