基于语音识别的动画人物唇同步系统

发布时间：2018-04-15 05:34

本文选题：隐马尔可夫模型 + 语音识别　；参考：《浙江大学》2008年硕士论文

【摘要】： 随着国内市场对动漫产品需求的不断增长,以及国家对动漫产业的大力扶持,我国的动漫产业进入了一个高速成长期。唇同步是动画制作中不可或缺的一个环节,但是传统的唇同步方法完全由人工来实现,非常繁琐,效率很低。并且唇同步的质量得不到保证,完全依赖于制作人员的经验。所以,我们需要一种技术来帮助动画制作人员摆脱这一繁琐的工作,同时提高唇同步的质量。本文应用基于隐马尔可夫模型的语音识别技术来解决动画的唇同步问题。通过对语音识别技术的研究,以及对唇同步问题的分析,我们找到了用语音识别技术解决动画人物唇同步问题的切入点:用语音识别技术来获取语音中每个字的时间信息。基于这个想法,我们设计了两种解决动画人物唇同步问题的方法。第一种方法通过直接应用识别算法来获取每个字的时间信息。第二种方法根据语音文本构建一个组合模型,然后在组合模型上应用Viterbi算法,在计算过程中获取每个字的时间信息。这两种方法都可以有效地解决动画的唇同步问题,但是和第一种方法相比,第二种方法有效地减少了计算量并且不存在识别错误的问题,因而在计算效率和唇同步的效果上都要优于第一种方法。最后,本文基于第二种方法实现了一个唇同步系统,并探讨了实现中的关键问题:如何根据系统特点确定模型结构,如何选择合适的识别单元,如何根据系统特点修改一般化的算法使其在我们的系统中正常的工作,如何对语音信号进行参数化。
[Abstract]:With the increasing demand of animation products in the domestic market and the strong support of the country to animation industry , China ' s animation industry has entered a high - speed long - term . The lip synchronization is an indispensable link in animation production , but the quality of lip synchronization is not guaranteed . It depends entirely on the experience of the producers . Therefore , we need a technique to help animation producers get rid of this complicated job while improving the quality of lip synchronization .

In this paper , the speech recognition technology based on hidden Markov model is applied to solve the lip synchronization problem of animation . Through the research of speech recognition technology and the analysis of lip synchronization problem , we find out the entry point for solving the problem of lip synchronization by using speech recognition technology : the speech recognition technology is used to acquire the time information of each word in the speech .

Based on this idea , we design two methods to solve the lip synchronization problem of cartoon characters . The first method obtains the time information for each word by applying the recognition algorithm directly . The second method can effectively solve the lip synchronization problem of the animation by applying the Viterbi algorithm on the combined model . However , the second method effectively reduces the calculation amount and does not have the problem of identifying errors . Therefore , the second method is superior to the first method in calculating efficiency and labial synchronization .

Finally , based on the second method , a lip synchronization system is realized and the key problem in the realization is discussed : how to determine the model structure according to the characteristics of the system , how to select the proper identification unit , how to modify the generalized algorithm according to the characteristics of the system makes it work normally in our system , and how to parameterization the speech signal .

【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2008
【分类号】：TP391.42

【共引文献】