Three Dimensional Modeling Of Speech Organs Based On MRI

Speech is an important way of information dissemination,and we have no idea how speech is produced? A lot of studies about how to communicate with people have already been carried out in human-computer interaction.However,few researches were investigated based on the Chinese Mandarin.Three-dimensional modeling of the larger Chinese pronunciation corpus is made by MRI in order to reveal the movement characteristics of the Chinese speech organs.In this paper,104 sets of different tomographic images of head were collected by magnetic resonance imaging(MRI),which contain three-dimensional information of vocal tract without the teeth in the oral cavity.The cone beam computed tomography(CBCT)was perfect for collecting the imaging of the bone structure.The data of the upper and lower teeth were superposed to the MRI images,which made up for the shortcomings of MRI without bone structure for the three-dimensional modeling of vocal tract.We considered the physiological boundaries of speech organs during tracing the organ contour to obtain more accurate organ model.At the same time,the main selection of the bone structure of the organ boundary as an anchor pointed to ensure consistency of organs under different pronunciation referring to the anatomical structure of the organ.Because some organs from the single point of view were difficult to ensure that the separation of organs boundary,we fuse organ contours of different viewpoint to construct the model of speech organs.We used a large number of points to trace the speech organs.The speech organs were represented by numerous vertexes to form point cloud data.These point cloud data were sorted in the certain rules,and then the points were connected with each other to form a 3D mesh of surface triangles to approximate the shape of the organs.The control parameters(two jaws,three tongues,two upper and lower lips,two soft palates,and three pharyngeal walls)were extracted by linear component analysis from the 3D models of the speech organs.The experimental results showed that using these control parameters could effectively describe the position and movement of the speech organs,in which root mean square errors were less than 0.2cm.