Department of Circuit Theory, Faculty of Electrical Engineering, Czech Technical University in Prague, Prague, Czech Republic.
Department of Circuit Theory, Faculty of Electrical Engineering, Czech Technical University in Prague, Prague, Czech Republic.
Comput Biol Med. 2023 Aug;162:107086. doi: 10.1016/j.compbiomed.2023.107086. Epub 2023 May 28.
Motor skills in children have traditionally been examined via challenging speech tasks such as syllable repetition, and calculating the syllabic rate using a stopwatch or by inspecting the oscillogram followed by a laborious comparison of the scores on a look-up table representing the typical performances of children of the given age and sex. As the commonly used performance tables are over-simplified to allow for manual scoring, we raise the question of whether a computational model of motor skills development could be more informative, and could allow for the automated screening of children to detect underdeveloped motor skills.
We recruited a total of 275 children aged four to 15 years. All the participants were native Czech speakers with no history of hearing or neurological impairments. We recorded each child's performance of/pa/-/ta/-/ka/syllable repetition. Various parameters of diadochokinesis (DDK; DDK rate, DDK regularity, voice onset time [VOT] ratio, syllable, vowel and VOT duration) were investigated in the acoustic signals using supervised reference labels. Female and male participants were analyzed separately by comparing younger, middle, and older age groups of children via ANOVA. Finally, we implemented a fully automated model that estimated the developmental age of a child based on the acoustic signal, and evaluated its accuracy using Pearson's correlation coefficient and normalized root-mean-squared errors (RMSEs).
The DDK rate reflected the ages of the children proportionally (p < 0.001). Other DDK parameters also showed strong sensitivity to age (p < 0.001), with the exception of VOT duration, which had a smaller effect (p = 0.091). The effect of age was found to be sex specific for the syllable length (p < 0.001) and DDK rate (p = 0.003). We observed that females spoke more slowly and had a longer VOT at preschool age (p < 0.001). The DDK rate obtained via the automated algorithm was strongly correlated with the reference (p < 0.001, Pearson's correlation coefficient of 0.97), with a low normalized RMSE of 3.77%.
As children develop their motor skills, they are capable of shortening the vowels to increase the rate of syllabic repetitions. The nonlinear development in childhood and adolescence, with a steady state in adulthood, follows a logistic function for the DDK rate. This study demonstrates that the development of motor skills can be examined sensitively and more appropriately by a fully automated noninvasive procedure that also accounts for the dispersion of values within age brackets.
传统上，儿童的运动技能是通过具有挑战性的言语任务来检查的，例如音节重复，以及使用秒表计算音节率，或者通过检查声谱图，然后通过查看代表给定年龄和性别的儿童典型表现的查找表来进行艰苦的比较。由于常用的表现表过于简化，无法进行手动评分，因此我们提出了这样一个问题，即运动技能发展的计算模型是否可以提供更有价值的信息，并可以实现对儿童的自动筛查，以发现发育不良的运动技能。
我们共招募了 275 名年龄在 4 至 15 岁的儿童。所有参与者均为母语为捷克语且无听力或神经损伤病史的儿童。我们记录了每个孩子的/pa/-/ta/-/ka/音节重复表现。通过使用监督参考标签，在声学信号中研究了各种言语的快动性（DDK；DDK 率、DDK 规则性、起音时间[VOT]比、音节、元音和 VOT 持续时间）参数。通过方差分析，分别对年轻、中年和老年儿童组的女性和男性参与者进行了比较。最后，我们实现了一个完全自动化的模型，该模型可以根据声学信号估计儿童的发育年龄，并使用 Pearson 相关系数和归一化均方根误差（RMSE）来评估其准确性。
DDK 率与儿童年龄成比例（p<0.001）。其他 DDK 参数也强烈反映了年龄（p<0.001），除了 VOT 持续时间（p=0.091），其影响较小。年龄对音节长度（p<0.001）和 DDK 率（p=0.003）的影响具有性别特异性。我们观察到，在学龄前，女性说话速度较慢，VOT 较长（p<0.001）。通过自动化算法获得的 DDK 率与参考值密切相关（p<0.001，Pearson 相关系数为 0.97），归一化 RMSE 低至 3.77%。
随着儿童运动技能的发展，他们能够缩短元音以增加音节重复的速度。儿童和青少年时期的非线性发展，在成年期达到稳定状态，DDK 率遵循逻辑函数。这项研究表明，通过完全自动化的非侵入性程序可以更敏感地检查运动技能的发展，该程序还考虑了年龄范围内值的分散性。
