使用語音處理量測朗讀流暢度之研究

Abstract

本研究目的為發展語音處理技術，用以輔助教師對兒童朗讀流暢度之評估。閱讀內容辨識是基於隱藏式馬可夫模型技術，輔以國語句法學之架構，以提升朗讀聲音的即時辨識率。流暢度的評量內容包含閱讀速度、字的正確率、及閱讀韻律等。閱讀速度是計算為每分鐘所朗讀的字數，正確度則是藉由語音辨識系統所求得，而閱讀韻律包含了三個部分：字或詞之間隔時間、音高以及重音。本研究實驗部分是藉由記錄不同閱讀能力的朗讀者所朗讀的國小四年級國語課文，比較各自的流暢度數據，以說明本研究所採用的評量方法。未來，根據評量數據所產生的朗讀流暢度指標，將可用來回授給閱讀者以提升其閱讀的成就。

The study investigates a signal processing technique for the assessment of oral reading fluency to assist children’s reading achievement. Reading voices recognition based on a Hidden Markov Model and the Mandarin Chinese syntax is used to improve the real time character recognition rate out of children’s reading voices. Fluency assessments were performed for reading speed, word accuracy, and prosody. Accuracy was estimated by a voice recognition system. Reading speed is defined as the number of characters read per minute. Prosody includes three parts: pause duration, pitch, and stress. Experiments were conducted to demonstrate the oral reading fluency measures derived from reading sounds of children with different fluency levels. Feedback instructions or indexes could be generated out of the oral reading fluency measures to children for improving their reading achievement.