论文部分内容阅读
为得到更加精确的声道共鸣特性,人们越来越关心说话时声道的结构及其形状变化。该文从MRI 3-D图像数据中提取了汉语普通话7个单元音[a]、[o]、[]、[i]、[u]、[y]、[e]的声道边缘、中线和矢量半径数据,并沿着声道中线从唇到声门等间距截取声道的36个截面,对于每个截面,根据矢量半径的大小使同一个位置的截面形状作线性过渡,从而建立用矢量半径驱动的汉语普通话立体声道模型。计算模型的共振峰并合成语音样本,在与自然语音对比的听辨实验中,该模型取得了较好的语音合成效果。
In order to get more accurate resonance characteristics of the channel, people pay more and more attention to the structure and the change of the shape of the channel when speaking. This paper extracts the channel edges of 7-unit Mandarin [a], [o], [], [i], [u], [y] and [e] of Mandarin Chinese from MRI 3-D image data, Midline, and vector radius data, and 36 sections of the channel are truncated along the centerline of the channel from the lip to the glottis. For each section, the cross-sectional shape of the same position is linearly transitioned according to the size of the vector radius so as to establish Mandarin Mandarin Stereo Model Driven by Vector Radius. Calculate the formant of the model and synthesize the speech samples, and get better speech synthesis results in the listening experiment compared with the natural speech.