Emotion, age, and gender classification in children's speech by humans and machines

In this article, we present the first child emotional speech corpus in Russian, called “EmoChildRu”, collected from 3-7 years old children. The base corpus includes over 20K recordings (approx. 30 hours), collected from 120 children. Audio recordings are carried out in three controlled settings by creating different emotional states for children: playing with a standard set of toys; repetition of words from a toy-parrot in a game store setting; watching a cartoon and retelling of the story, respectively. This corpus is designed to study the reflection of the emotional state in the characteristics of voice and speech and for studies of the formation of emotional states in ontogenesis. A portion of the corpus is annotated for three emotional states (comfort, discomfort, neutral). Additional data include the results of the adult listeners’ analysis of child speech, questionnaires, as well as annotation for gender and age in months. We also provide several baselines, comparing human and machine estimation on this corpus for prediction of age, gender and comfort state. While in age estimation, the acoustics-based automatic systems show higher performance, they do not IThis is the unedited author proof of the accepted article. ∗Corresponding author Email addresses: [email protected] (Heysem Kaya), [email protected] (Albert Ali Salah), [email protected] (Alexey Karpov), [email protected] (Olga Frolova), [email protected] (Aleksey Grigorev), [email protected] (Elena Lyakso) Preprint submitted to Computer Speech and Language July 17, 2017 reach human perception levels in comfort state and gender classification. The comparative results indicate the importance and necessity of developing further linguistic models for discrimination.