This paper introduces the extended description of a database that contains emotional speech in the Russian language of younger school age (8–12-year-old) children and describes the results of validation of the database based on classical machine learning algorithms, such as Support Vector Machine (SVM) and Multi-Layer Perceptron (MLP). The validation is performed using standard procedures and scenarios of the validation similar to other well-known databases of children’s emotional acting speech. Performance evaluation of automatic multiclass recognition on four emotion classes “Neutral (Calm)—Joy—Sadness—Anger” shows the superiority of SVM performance and also MLP performance over the results of perceptual tests. Moreover, the results of automatic recognition on the test dataset which was used in the perceptual test are even better. These results prove that emotions in the database can be reliably recognized both by experts and automatically using classical machine learning algorithms such as SVM and MLP, which can be used as baselines for comparing emotion recognition systems based on more sophisticated modern machine learning methods and deep neural networks. The results also confirm that this database can be a valuable resource for researchers studying affective reactions in speech communication during child-computer interactions in the Russian language and can be used to develop various edutainment, health care, etc. applications.

The paper presents a description of the methodological approach for assessing the formation of the emotional sphere of children aged 5-16 years with typical and atypical development. The purpose of the Child's Emotional Development Method (CEDM) is to assess the emotional development of the children by determining the ability to express their own emotions, emotions adequacy, and recognition of the emotional states of others. The approach is based on adapted methods and scales, assessment criteria and methodological approaches tested in a number of studies, depending on the age of the children, their developmental characteristics, language and cultural characteristics of the country of residence of the child. The approach includes two blocks - information about the development of the child obtaining from parents / legal representatives; methods for children testing, including interview methods, psychological tests, play situations. The assessment uses a score on the Likert scale. A pilot approbation of the methodological approach was carried out on 30 children aged 8-16 years - typically developing (TD), with autism spectrum disorders (ASD) and Down syndrome (DS). The results of the study formed the basis for the creation of the Scale of Emotional Sphere Development of Children and Adolescents.

The aim of the study is to identify the speech features of children with mild intellectual disabilities (ID) vs typically developing children (TD) based on the complex approach that includes perceptual, phonetic, linguistic, and spectrographic methods. The participants of the study were 60 Russian-speaking children aged 5–7 years: 20 children with ID and 40 children with TD. The peculiarities of the structure of dialogues between adult and children with ID and TD were described. Perceptual study showed that native speakers recognize the meaning of words of children with ID worse vs TD children. Listeners classify the emotional speech of children with ID with higher accuracy than the emotional speech of TD children. Spectrographic analysis revealed higher values of pitch and the duration of words and vowels in words of spontaneous and emotional speech of children with ID vs TD children.

Cross-cultural study that avoids bias in the formulation of emotion recognition models is indispensable to address the challenges in facial emotion classification for children, for it being a relatively difficult task. With evidences from literature for the hybrid feature extraction approaches to improve the image classification accuracy, this research work focuses on developing a hybrid framework for emotion recognition from facial images of Tamil and Russian children. The dataset is audio video recording of 28 Tamil speaking and 64 Russian speaking children. The data is collected in a controlled environment and labelled by experts. Traditional features like Grey Level Cooccurrence Matrix (GLCM) and facial landmark are extracted and are fused with You Look Only Once (YOLO V5) features. While facial landmarks and GLCM provide useful information about the facial expressions and texture of the image, YOLO V5 being a single-stage object detector makes the hybrid model super-fast and achieve high accuracy in detecting small objects and in low-light settings. The different classifiers that includes KNN, SVM, Random Forest, XGBoost, and Multilayer Perceptron, is employed yielding the accuracy result of 93%, 86%, 89%, 88%,and 90%. The use of majority voting ensemble of heterogeneous classifiers for the final prediction strengthens the model further, yielding an accuracy as high as 96% for the custom cross-cultural dataset, consisting of facial images of Russian and Indian children. This is consistent with the results obtained for Indian and Russian datasets. Further, the ablation study unveils the effect of feature fusion in boosting the performance and the dominance of YOLO V5 features over the other two.