Abstract
隨著網際網路的蓬勃發展，許多網際應用程式相繼應運而生，人們在使用這些應用程式時往往需要一個虛擬的替身，通常我們利用“使用者名稱“或是“使用者圖像“等方式來代表。近年來利用三維虛擬人類作為即時互動界面的應用將隨著軟硬體平台的進步而逐漸增加，在利用虛擬人類作為與他人互動的虛擬代理人時，我們不希望看到的是一個只有肢體動作但面無表情的冷面替身，因此虛擬人類溝通時的臉部表情以及說話時唇形的變化佔了很重要的地位。為了能呈現出人類說話時逼真的唇形，電腦唇形動畫研究的重點大部分都著重於語音和嘴唇的同步上，希望透過合成的語音和嘴唇的動作來模擬理想中虛擬人類的表情。不過目前在這方面的研究大部分都是以英文為主。在本論文中，我們提出一個國語的唇形動畫系統架構，並依照此架構實際的建構出一套唇形動畫系統M-Lips (Mandarin LipAnimation System)。這個系統利用中文文字當作輸入，並將國語文字注音音節對應至中文唇位表，以產生相對應的唇位號碼。我們所使用的中文唇位表是依據中文文字語音的特性加以分析出來，在經過整理後列出了十二種中文標準唇位。在嘴唇模型的塑形上我們使用了七條伯齊耳造型曲線/ 曲面，並利用嘴唇的對稱性來塑形出完整的唇形。接著我們嘗試利用具有物理性質的“動態節點位移演算法“計算出唇形變化的軌跡，來移動我們的嘴唇模型，以達到唇形動畫的效果。本系統經與中文語音合成系統整合後能按文發音且音唇之同步情形良好。Recently, the popularity of Internet has promoted thedevelopment of Internet-based applications.People who use theseapplications always need to have a virtual representative suchas "user ID" or "user icon" etc..Due to the advance of softwareand hardware technology, the use of 3D virtual human as anavatar in the application has been studied extensively. Whileapplications use virtual humans as virtual representatives,people expect both the movements of bodies and the facialexpressions. The lip movement during speaking plays a veryimportant role incommunication. There are many researches instudying the synchronization between lip shape and speech.However, these researches mostly focus on English.In thisthesis, we proposed a software system architecture forgenerating mandarin lip animation. Besides, we implement asystem called M-Lips (Mandarin Lip Animation System) based onour architecture. First, a Chinese viseme table is analyzedaccording to the Chinese phonetic characteristics and it hasbeen classified into 12 standard Chinese visemes.Then an inputtext is decomposed into syllables for retrieving thecorresponding viseme numbers in the "syllable to viseme" mappingtable. Moreover, we use cubic Bezier curve to model the lip andthen calculate the nodal trajectory of lip by using dynamicnodal displacement algorithm. Finally, we integrated M-Lips withthe Chinese speech synthesizer.The experiment systemsatisfactorily reads the given text and converts it to speechwith lip animation.