TY - GEN

T1 - Assessment of Children’s Ability to Manifest Emotions in Facial Expressions, Voice and Speech by Humans, Automatic, and on a Likert Scale

AU - Ляксо, Елена Евгеньевна

AU - Фролова, Ольга Владимировна

AU - Матвеев, Антон Юрьевич

AU - Николаев, Александр Сергеевич

AU - Ruban, Nersisson

N1 - Conference code: 26

PY - 2024/11/25

Y1 - 2024/11/25

N2 - The goal of the study was to assess children’s ability to manifest emotions in facial expressions and speech by humans, automatic and using Likert scale scores. To achieve this goal, two studies were conducted. The first study performed a perceptual and automatic analysis of the emotions “joy - neutral - sadness - anger” in typically developing (TD) children; the second - compared emotion recognition in four groups of children - TD, autism spectrum disorders (ASD), intellectual disabilities (ID) and Down syndrome (DS) by expert and automatic, and analyzed Likert scale scores for completing test tasks. The participants of the study were 110 children aged 5 - 16 years, 18 adults. The original dataset containing video and audio fragments of children’s emotional states was used. Experts recognize the emotions of children from all groups by video and speech more accurately than automatic classifications, with higher UAR values for TD children by audio and video in a perceptual experiment and by audio in the automatic classification of emotions. Differences in the classification accuracy of emotions in children with ASD, ID, and DS were identified. Sadness and anger states are automatically classified poorly by audio and video in children with ASD, ID, and DS. The novelty of the results lays in the obtaining normative data on the recognition of emotions in TD children and in comparative data on groups of TD children, children with ASD, ID, DS.

AB - The goal of the study was to assess children’s ability to manifest emotions in facial expressions and speech by humans, automatic and using Likert scale scores. To achieve this goal, two studies were conducted. The first study performed a perceptual and automatic analysis of the emotions “joy - neutral - sadness - anger” in typically developing (TD) children; the second - compared emotion recognition in four groups of children - TD, autism spectrum disorders (ASD), intellectual disabilities (ID) and Down syndrome (DS) by expert and automatic, and analyzed Likert scale scores for completing test tasks. The participants of the study were 110 children aged 5 - 16 years, 18 adults. The original dataset containing video and audio fragments of children’s emotional states was used. Experts recognize the emotions of children from all groups by video and speech more accurately than automatic classifications, with higher UAR values for TD children by audio and video in a perceptual experiment and by audio in the automatic classification of emotions. Differences in the classification accuracy of emotions in children with ASD, ID, and DS were identified. Sadness and anger states are automatically classified poorly by audio and video in children with ASD, ID, and DS. The novelty of the results lays in the obtaining normative data on the recognition of emotions in TD children and in comparative data on groups of TD children, children with ASD, ID, DS.

KW - Emotional State

KW - Perceptual and Automatic Recognition ·Children

KW - Video

KW - Audio Modalities

KW - Likert Scale Scores

KW - Perceptual and Automatic Recognition

KW - Children

UR - https://www.mendeley.com/catalogue/2b2c9fda-1d1b-3b19-b539-1a9ac9448690/

U2 - 10.1007/978-3-031-77961-9_21

DO - 10.1007/978-3-031-77961-9_21

M3 - Conference contribution

SN - 978-3-031-77960-2

T3 - Lecture Notes in Computer Science ( Lecture Notes in Artificial Intelligence)

SP - 281

EP - 294

BT - 26th International Conference, SPECOM 2024, Belgrade, Serbia, November 25–28, 2024, Proceedings, Part I

A2 - Karpov, Alexey

A2 - Delić, Vlado

T2 - 26th International Conference on Speech and Computer

Y2 - 25 November 2024 through 28 November 2024

ER -