Construction And Control Of Three Dimensional Geometric Articulatory Model

Articulatory synthesizers imitate the articulatory and aerodynamic process of speech production.The current work describes an attempt to construct a more accurate articulatory model to approximate the morphological properties of speech apparatus.This may lead to better models for articulatory synthesizers.There are two main modeling strategies,physiologic modeling,and geometric modeling.This paper focuses to build a vocal tract model by the side of pure geometry.This paper focuses to build a vocal tract model by the method of pure geometry.The geometry model is different from the neurophysiologic model,it do not need to consider the effection of muscles.The advantages lies in the following: less amount of calculation,real-time performance.This study describes a novel method of constructing a geometric articulatory model based on MRI(Magnetic Resonance Imaging)and CBCT(Cone beam C T)data by taking the physiological boundaries of speech apparatus into account.MRI is a new and very useful method to collect clear contours of the vocal tract organs.At the same time,it is harmless to people,so it becomes more and more popular in speech synthesis field.Due to the jaw and palate are bony structures,which cannot be imaged by MRI directly,so we acquired the CBCT data of vocal tract.From the database,we can extract a accurate vocal tract shape,and from the extracted shapes of different vowels,we can get the law of shape variety.The research is significant in many fields,such as speech production,speech analysis,speech teaching and the visualization of the whole vocal tract.We choose 104 pronunciation data of one subject to do the research,which is composed of data acquisition and preprocessing,data annotation and 3D mesh modeling,data analysis and validation evaluation,collision detection and response.Results of linear component analysis method show that every speech organ can be described well by less than three parameters,and the accumulated contribution rate is more than 80%.After the reconstruction,the root-mean-square error is less than 1.0 mm.This paper provides a new method to annotate and construct 3D vocal tract organs by taking the physiological boundaries of speech apparatus into account.Two improvements have been made to the modeling process: i)Images taken from different viewpoints are combined to improve the accuracy of outline annotation.ii)Speech organs' meshes are modeled with reference to the anatomical structures.This has two advantages according to the existing method: the integrity of the speech organs and the relativity of the special feature points of the organs.In the end,we constructed a 3D vocal tract geometry model based on Chinese database.It is significant to the applications about Chinese speech,such as language teaching,vocal tract visualization,the popularization of Chinese and so on.