Synthesis of Speech Animation Based on Viseme Analysis

• Main Authors: Kun-syong Jiang, 江坤雄
• Other Authors: Yih-Kai Lin
• Format: Others
• Language: zh-TW
• Published: 2010
• Online Access: http://ndltd.ncl.edu.tw/handle/94938321067338443855

碩士 === 國立屏東教育大學 === 資訊科學系 === 98 === The research is expected to analyze the change of the lips by the process of the real person pronunciation,and arrange with the recognition of the speech to form the output the lips animation of real facial. First,we took the image of the real person pronunciation with the speech , and then separated the sound and the image . Through the training of Hidden Markov Model,the part of the sound would produce several models which would be recorded in order to synchronize . In the part of the image,we use the facial detected pattern to find the faces . Because we set up the shoot of the front face,we could acquire the position of the lips according to the proportion and find the red parts by the process of the image to acquire the characters of each pronounced lips , and combine the sound and the image to produce synchronization model . First,the users spoke the speech , and then we took the audio frequency to compare with the HMM models which had been established to acquire the model with the highest probability in Hidden Markov Model . The model would make the nearest audio frequency with the pronunciation information compare with and search the database of the lips which had been established to produce the output of the synchronization.