本研究目的為使用寵物型機器人,在和使用者的互動朗讀過程中,評估閱讀者文句節奏狀況,並給予回饋進而幫助閱讀流暢度的改善。辨識朗讀者節奏,必須先利用語音端點偵測技術,找到字的端點。本研究機器人的端點偵測系統,是使用過零率、能量、頻譜、自相關函數等特徵,訓練隱藏式馬可夫模型之分類器。互動的機制上,在機器人上使用兩個垂直方向的馬達,發展出具節拍器功能之尾巴,用以給予使用者閱讀節奏之引導。 為了評估讀者和機器人相互影響關係以及讀者節奏狀態。以句子為單位,定義字相位模型,並用此定義同步參數以及節奏參數。其中同步參數用以評估讀者和節拍器的相互關係,節奏參數用以定義句子結構以及評估讀者節奏。評估使用者節奏參數,若其沒有達到指定的朗讀節奏,在下一次的朗讀過程,節拍器回授使用者一個更顯著差異的時間差,使之更具朗讀節奏。而時間的改變量也將根據同步參數不同而有所調整。 本研究實驗部份,首先藉由分析有節拍器引導與否,顯示尾巴引導機制的效果。接著藉由重複的引導過程,觀察使用者朗讀狀態的改變。最後用隨機決定的朗讀文句,觀察使用者受節拍引導的影響。未來,可用以陪伴幼童讀書以及幫助其提升對文章節奏的熟悉度。
In this study, the pet-like dog that interacts with reader for enhancing the reading cadenced is being developed. Acoustic behaviors of voice signals are recorded to analyze the condition during interaction. The reader reading rhythm is found by endpoint detector. The endpoint detector is a classifier and based on the Hidden Markov Model. It is trained by nonverbal auditory cues, such as zero-crossing rate, energy, spectrum, and autocorrelation. The spectrum and autocorrelation are chosen to recognize the periodic signal. To provide guidance of oral reading, the robot’s tail is developed into a two directions metronome by the two perpendicular motors. The goal is to classify reading states of the user, so the phase model of the words’ period in a sentence is defined, then the synchronization and rhythm parameter are defined by word phases. The synchronization parameter characterizes the users’ response models with respect to the tail input. And the rhythm of the sentence is characterized by the rhythm parameter. Measuring the rhythm parameter, if the rhythm parameter is far from the commanded rhythm, a new pace is then set up to control the metronome for the sentence reading guidance. The guidance rule for a specific user considers not only the rhythm parameter difference but also the synchronization parameter of the user. Experiments were conducted to demonstrate the effect of guidance on oral reading performance. Rhythm parameters being controlled to approach better reading fluency is observed under the proposed guidance rule.