Modeling Of3D Geometry Vocal Tract In The Procession Of Speech Production

Magnetic Resonance Imaging (MRI) is a new and very useful method to collectclear contours of the vocal tract organs. At the same time, it is harmless to people, soit becomes more and more popular in speech synthesis field.We have a database of8Chinese subject, each one has10vowel data.From the database, we can extract anaccurate vocal tract shape, and from the extracted shapes of different vowels, we canget the law of shape variaty. The research can get the result of visualization of thewhole vocal tract, which is significant in many fields, such as speech teaching, speechproduction, speech analysis.This paper focuses to build a vocal tract model by the side of pure geometry. Thegeomery model is different from the neurophysiological model, it do not need toconsider the effection of muscles. The advantages lies in the following: less amount ofcalculation, real-time performance.We choose one of all the eight subjects to do theresearch, which is composed of pre-processing, tooth filling, annotation,3D meshingand analysis.Theanalyse method is Liner Component Analysis(LCA).The result shows that, every speech organ can be described well by less than threeparameters, the accumulate contrabution rate is more than88%. After thereconstruction, the average error is less than1.0mm.This paper provides a new method to annotate and construct3D vocal tractorgans, which has two advanges according to the existing method:The first one is the integrity of speech organs. For example, thetongue shapeconstructed by the existing method has a deficiency on tongue root and the part whichconnects tongue and jaw, which vary during the speaking. So, this will lead errors tothe result, which can be avoided by our method.The second one is the relativity of the special feature points on the organs. Suchasteeth tip, tongue tip, tongue root. We seperate the tongue surface andother parts ofthe3D tongue shapes, because their variaty level are different. However, thetraditional method treats them equally. So, our result is better on the shapes relativity.This is the first test to construct a3D vocal tract geometry model based onChinese. It will provide theoretical basis to the applications such aslanguage teachingon Chinese and the popularization of Chinese.