论文部分内容阅读
言语产生是一个非常复杂的认知过程,这个过程主要由概念形成、言语组织和发音运动三个部分组成。神经建模是研究脑神经通路的一种方法,主要作用是分析大脑中各个功能区的分工及信息的传递。文章以关于言语产生的神经模型DIVA为例,介绍言语产生神经建模的基本原理,同时指出DIVA模型存在的缺陷。我们提出一种新的言语神经建模方法,使用末梢生理发音模型生成语音数据,利用神经网络建立神经指令和语音参数之间的关系,并且对运动神经指令进行最优化,利用优化阶段的准则缩减冗余度,实现从发音目标到声道形状的产生。
Speech production is a very complex cognitive process, which is mainly composed of three parts: concept formation, speech organization and pronunciation movement. Neural modeling is a method of studying the neural pathway of the brain, the main function of which is to analyze the division of labor and the transmission of information in various functional areas in the brain. Taking DIVA, a neural model of speech production as an example, this paper introduces the basic principles of speech-generated neural modeling and points out the defects of DIVA model. We propose a new speech neural modeling method to generate speech data using a peripheral physiological speech model, establish the relationship between neural instructions and speech parameters using neural networks, and optimize the motoneuronal instructions using the guidelines of the optimization phase Redundancy, from the pronunciation of the target to achieve the shape of the channel.