An articulatory model of standard Chinese using MRI and X-ray movie

Abstract

To better understand speech production from the phonological inputs to articulatory movements and then to acoustic outputs, it is important to establish an elaborate articulatory model of the vocal tract. This paper has explored the articulatory mechanism of speech production in Standard Chinese and developed a geometric articulatory model in both the visual and acoustic modalities. This model was based on the data of MRI images and X-ray movie, with the former providing detailed volumetric information of the vocal tract, and the latter the dynamic information of articulation. In this model, the seven articulators have been studied and modeled, including the hard palate, pharynx, jaw, lips, velum, tongue, and larynx. The tongue is modeled as two parts: tongue tip and tongue body, thus reducing the necessary number of parameters. The relation between larynx height and fundamental frequency in regard to the four tones is also modeled. These two improvements on tongue and larynx modeling have contributed new ideas to the articulatory modeling of Standard Chinese. This model can serve as a research tool for linguists, phoneticians, and speech engineers, and can be used in parameter speech synthesis, virtual speaker, and visual assistant speech training of Standard Chinese. 提要: 爲了更好地理解言語的產生過程,即如何從音位輸入到調音器官的動 作、再到聲學輸出,需要建立詳盡的聲道調音模型。本文是對漢語普 通話言語產生中調音機制的探索與研究,建立了一個具有視覺和聲學 輸出的幾何調音模型。該調音模型的數據源自於聲道的磁共振圖像和 X 光錄影,前者主要提供聲道的立體形狀,後者提供調音的動態過程。 在這個模型中,聲道被分解為七個調音部位進行研究:硬齶、喉腔後 壁、下頜、雙唇、軟齶、舌頭和喉管。其中創新性地,舌頭又被分為 舌體(相對簡單)和舌尖(更為靈活)兩部份,從而簡化所需要的參 數;另外,普通話四聲中基頻高低與喉管上下高度的關係也加入到模 型中。該模型可以作為研究工具服務於語言學、語音學和言語工程, 並可用於語音參數合成、虛擬說話人、普通話輔助教學等領域。