腭裂代偿性发音的声学特征分析、建模和客观评估方法的研究

项目名称： 腭裂代偿性发音的声学特征分析、建模和客观评估方法的研究

项目编号： No.60875014

项目类型： 面上项目

立项/批准年度： 2009

项目学科： 金属学与金属工艺

项目作者： 赵庆卫

作者单位： 中国科学院声学研究所

项目金额： 33万元

中文摘要： 本研究在对腭裂患者发音特性进行深入调研的基础上，利用声学分析法对代偿性发音进行研究，提出了能够对腭裂代偿性发音进行客观评价的声学特征及相应的评估方法，有利于腭裂患者的治疗和康复。主要研究成果如下： 1. 在标注中发现，代偿性发音(声门顿音)大致可分为两类，两类之间的区别主要在于患者在何时进行爆破的释放，在声谱图上就表现为是否存在明显的辅音段。 2. 针对第一类声门顿音会出现塞音段丢失的现象，提出基于听觉滤波器组的方法对塞音爆破丢失的现象进行检测。该方法采用类听觉的前端处理环节，并对处理后信号在其各子带内分别计算塞音除阻过程中的能量变化率；最后检验与主观判听的一致性。 3. 由于声门顿在声门（或喉）处成阻，使得口腔内成阻不再必须。降低的软腭容易在后续的元音段中持续，导致元音的鼻化增加。因而提出采用1/3倍频分析对过高鼻化进行检测。 4. 从发音的角度看, 更低的成阻位置对应于更长的前腔。因而，提出采用Spectral moments 对发音时成阻位置进行检测。 5. 针对声门顿的发音特点，提出采用LF声门波模型的方法对腭裂患者声带异常程度进行检测。

中文关键词： 声门顿;喉化;高鼻化;声学分析;腭裂

英文摘要： Based on investigation of pronunciation characteristics of cleft palate patients, the method of acoustic analysis is used in patients with cleft palate compensatory articulation. The acoustic feature and corresponding objective evaluation method on compensatory articulation is put forward,which is helpful for the treatment and recovery of the cleft palate patients. The main research results are as follows: Using spectrogram analysis, two types of compensatory articulation (glottal stops) were revealed. The primary difference between these two cases is the absence/presence of common consonants, which is determined by the burst-release time. A method, based on the gammatone filter-bank, is proposed to detect the absence/presence of plosives in glottal stops. The auditory filter-bank is used as front-end processor. Then, the energy variation during the release phase of stops within each sub-band is estimated. Next, the average energy variation within each sub-band is compared between the cleft palate group and post-palatoplasty group using t-test. Finally, Logistic regression is used to assess the consistence between the proposed method and subject evaluation. When glottal stop is used as a substitute for high pressure consonant, the air stream is blocked at the glottis. In this case, velopharyngeal closure is not necessary. Hence, if the glottal stop is produce with a low velum, one can expect that it may not rise during the production of the following vowel. In this work, one-third-octave spectra analysis is used to evaluate the hyper-nasality associated with glottal stops. The low obstruction point corresponds to a long front cavity. Therefore, we propose to use spectral moments as the measurement of the front cavity length. Due to the laryngealization in cleft palate speakers, source excitation is fitted to a standard glottal model (the Liljencrants- Fant Model) in order to acquire the abnormal vibration of vocal folds, such as a low OQ (open quotient) and spectral tilt.

英文关键词： Glottal stop;Laryngealization; Hypernasalitys;Acoustic Analysis;Cleft Palate