Synthesis Of Visual Co-articulation For Uyghur Language

Expressive visual speech synthesis is now playing an important role in the field ofcomputer graphics and also a hot issue.It has great research value whether in theoryor in practical applications. Currently, visual speech synthesis studies at home andabroad focus more on English, Chinese and other widely used languages and havemade great achievements, many of which have been applied in real life. However,studies on minority languages are still relatively few in which Uyghur speechsynthesis has already achieved some results, but the study on the visual speech of thislanguage is still empty. So this paper aims to study techniques and methods for visualco-articulation in visual speech synthesis of Uyghur language.Of the studies on those languages with more mature visual speech synthesistechnology, technical research on English and Chinese, the commonly used languages,are focused on, based on which the thesis uses method of dynamic visual model basedon weights to synthesize visual co-articulation for Uyghur language i.e. the animationparameters of Uighur co-articulation. Firstly, the thesis analyzes differences on thesynthetic methods of visual co-articulation and introduces the establishment of theUighurs voice and video database. Secondly this thesis introduces the dynamic visualmodel based on weights and analyzes impacts of various parameters in the model onthe final synthesis of dynamic viseme followed by a proposal to use the method ofmanual adjustment combined with gradient descent algorithm to obtain the modelparameters. Then a brief comparison is presented on the synthesis effect respectivelybased on control of model parameters and interpolation method based upon staticviseme. And then, after a grouping of those FAP parameters in control of visemechange, the thesis chooses the representative ones and uses the model parameters forthem to classify32Uighur letters. Finally, the previous classification is used for thesynthesis of the dynamic viseme of Uyghur Syllables and with an experimental comparison on the synthesis results, the thesis summarizes the impact of differentclassifications on synthesis effects.