語音辨識軟體回饋對發音學習成效之影響研究

Abstract

本研究旨在了解語音辨識軟體之回饋對大學生學習發音的效用。研究中邀請了二十三位來自台灣師範大學大一英文課的學生參與本研究中的實驗組，另有二十位來自同樣大一英文課的學生作為控制組。兩組學生皆需接受有二十句包含易混淆母音句子的前測，並將學生朗讀的內容錄音。接著實驗組的學生必須接受為期五週的發音練習，期間使用軟體MyET。控制組的學生在這五週期間則不需參與任何發音訓練。本研究並採用了回饋單、問卷和訪談。此外，學生使用軟體作發音練習的過程亦由螢幕擷取軟體Camtasia Studio所記錄，學生練習的語音部分亦會同時被錄製，以作為觀察用。整個發音訓練結束後，實驗組和控制組學生皆需參加後測和全新的歸納測驗（Generalization test）。最後所有的錄音檔案交由三名評分老師評分，以了解兩組學生在測驗中的表現是否有差異。 研究的結果顯示實驗組的學生經過五週的發音練習後，在後測和歸納測驗中，在單音的表現和音調的部分有顯著的進步。在評估不同類型的回饋中則發現隨著練習時間的增加，學生對各種回饋的信心也有變化。在研究結束前，學生認為發音回饋、發音診斷、分數、語調回饋和語調波紋圖對發音學習最為有效，而聲譜圖、節拍回饋和音量回饋則被評鑑為無效的回饋。因此，從研究結果中得知學生確能從一些特定的回饋中改善他們的發音或語調。更重要的是，學習者在練習的過程中察覺到自己的發音問題，並發展出一些策略來改進。 最後對於未來的相關研究提出一些建議。首先，發音練習開始前應教導學習者如何解讀不同的回饋類型。此外，發音練習所使用的句子應該要儘可能地簡單，才不致使學習者產生閱讀障礙。至於往後相關研究則建議採用更多的受試者，較長的訓練時間和在各個測驗中使用一樣數量的句子。另外，也可增加後續測驗來檢測訓練成果持續的時間。而若是能再研究中使用另一個有類似回饋的語音軟體來做比對，那麼各種不同類型的回饋效用則能有更肯定的佐證。The current study aims to investigate the impact of ASR software feedback on EFL college students’ pronunciation learning. Twenty-three students from Freshman English classes in NTNU attended the study as the experimental group and twenty as the control group. Both groups took a pre-test containing twenty assigned sentences with confusing vowels for Chinese speakers and their speech productions were simultaneously recorded. Following the pre-test, the participants of the experimental group received a five-week pronunciation training with the software MyET, whereas the control group did not. The participants of the experimental group also had to complete feedback forms, questionnaires and interviews in the study. Besides, training observations were completed by using screen recording software, Camtasia Studio. After the training session, the participants of both the experimental group and the control group took a post-test and a generalization test. Three raters then assessed the collected recordings and gave a segmental score and a prosodic score to each of them. The findings of the study revealed the five-week pronunciation training session successfully resulted in the experimental group’s significant improvement in segmental accuracy and prosodic productions in the post-test and in the generalization test. In evaluating the different types of feedback, practice time was found to have affected the participants’ confidence in the feedback types. At the end of the study, the participants considered feedback on pronunciation, pronunciation diagnosis, scores, feedback on intonation, and pitch contours more effective in pronunciation learning than spectrograms, feedback on rhythm and feedback on volume. Learners, therefore, benefited from some of these feedback types in improving either their segmental accuracy or prosodic productions. Most important of all, learners could perceive their pronunciation errors and had cultivated certain strategies to correct these errors after the training. It is then suggested that extra instructions on how to interpret the feedback be given to learners before the pronunciation training starts. Besides, the pronunciation task should be as simple as possible to ensure that learners have no difficulty reading the sentences. For future studies, a larger size of participants, longer training session, and equal number of sentences with proximate sounds in the pre-test, post-test, and the generalization test are recommended. A follow-up study to examine the duration of the training effect is also practical. Moreover, another ASR software with similar design may be employed in the same study in order to verify whether feedback of the same types indeed work for language learners even when different software is used.