Study on Natural Voice Synthesis based on Analysis of the Articulation Organ.

自然语音合成是语音处理系统和人工发音系统开发的重要课题。声带和声道是发声系统的两个基本器官。语言的自然性，如重音和语调，取决于声带对音高的控制。声道控制在发音中的重要性是不言而喻的。我们通过关节运动的测量等多种方法对关节系统进行了分析。PARCOR分析、咽喉阻抗测量和使用磁共振（MRI）的声道成像。在本项目中，后两种方法进行了讨论。(1)A研制了一种喉道阻抗计，可分别测量电阻和电容分量。音高与阻抗之间的平稳特性可以用三阶多项式函数来近似。利用FFT分析了阻抗与基音周期的动态关系，从阻抗的变化来估计基音周期. S ...更多信息 在听力测试中，用估计的音高合成与原始元音一致的元音。在连续元音的情况下，可以通过补偿由声音变化引起的阻抗分量来估计音高。对于大部分阻力变化不伴随音高变化的被试，容量变化对音高估计是有效的。据推测，这种差异应该取决于发音的parsonality。（2）利用MRI测量声带和鼻腔的三维形态，特别是对低水合物含量的牙冠，在其上覆盖一层薄的牙冠板，以获得图像。使用这种方法，磨牙之间的距离可以测量精度低于0.4 mm。在发音日语元音和摩擦音（/s/，/sh/，/shu/）时，可以通过应用三次卷积方法对3D图像提取声道和鼻腔的任意部分。结果表明，在咬合过程中，牙冠的位置和鼻旁腔的容积存在着不同的部位。少

Natural voice synthesis is an important project for developing the voice processing system and the articulation prosthesis system. The vocal cords and the vocal tract are two elementary organ in the voice articulation system. Naturality of language such as accent and intonation depend on the pitch control in the vocal cords. Significance of the vocal tract control in articulation is self-evident. We have analyzed articulation system by several methods such as measurement of articulation movement. PARCOR analysis, measurement of throat impedance and vocal tract imaging using magnetic resonance(MRI). In this project, later two methods are discussed.(1)A throat impedance-meter was developed for measuring resistance and capacitance components separately. Stational characteristics between vocal pitch and impedance could be approximated by a third order polynomial function. Dynamic characteristics between vocal pitch and impedance was analyzedby FFT to estimate pitch from impedance change. S … More ynthesized vowels using estimated pitch accorded with original vowels in hearing test. In the case of sequential vowels, vocal pitch could be estimated by compensating impedance component arising from change of the vocal sound. For most part of the subjects whose resistance change did not accompanied with vocal pitch, capacity change was effective for pitch estimation. It was supposed that this variance should depend on parsonality in the articulation.(2)Three dimensional shapes of vocal tract and nasal cavity were measured using MRI.Especially, image of dental crowns containing low hydrate content was obtained by covering them with thin dental crown plate. Using this method, distance between molars could be measured in accuracy below 0.4 mm. Arbitrary sections of the vocal tract and nasal cavity could be extracted by applying cubic convolution method to 3D images during articulating Japanese vowels and fricatives(/s/, /sh/, /shu/). It was revealed that there existed parsonal differences in the situation of the dental crowns and volume of the paranasal cavities during articulation. Less