Automatic Evaluation of Soft Articulatory Contact for Stuttering Treatment
Automatic Evaluation of Soft Articulatory Contact for Stuttering Treatment
复制标题
DOI：
10.21437/interspeech.2018-2544
复制
发表时间：
2018-09
期刊：
影响因子：
--
通讯作者：
K. Ochi;K. Mori;Naomi Sakai
中科院分区：
文献类型：
--
作者：
K. Ochi;K. Mori;Naomi Sakai
我们描述了一种从具有软发音接触的辅音开始自动辨别和评估发声的新方法，该方法用于治疗口吃和正常发声。软关节接触经过训练可以放松咬合架并消除口吃期间发生的硬接触。我们使用与声学特性和语音质量变化相关的特征，假设初始辅音的发音运动减慢和发声肌放松与软发音接触同时发生。实验评估的结果表明，当声学特征与梅尔频率倒谱系数（MFCC）的一阶导数峰值相关时，与咬合架运动减慢相对应，可以获得高精度。声音质量的特征对分类的影响很小。
We describe a new method for the automatic discrimination and evaluation of phonation beginning with a consonant with soft articulatory contact, which is used in the treatment of stuttering, and normal phonation. Soft articulatory contact is trained to relax articulators and remove hard contacts that occur during stuttering. We use features related to the changes in acoustic characteristics and the voice quality under the hypothesis that the slowing down of articulatory movement of the initial consonant and the relaxing of phonatory muscles co-occur with soft articulatory contact. The results of an experimental evaluation showed that high accuracy was obtained when acoustic features were related to the peaks of the first derivative of the mel frequency cepstral coefficients (MFCCs) corresponded to the slowing down of the movement of the articulators. The features of vocal quality only slightly contributed to the classification.