This paper explores the *text-based emotion prediction* problem, aiming to classify the emotional content of sentences in children's fairy tales for text-to-speech synthesis. The authors use supervised machine learning with the SNoW learning architecture to classify emotional versus non-emotional content. Initial experiments on a dataset of 22 fairy tales show promising results, outperforming a naive baseline and a bag-of-words (BOW) approach. The study also discusses a tripartite model that considers emotional valence and explores different feature sets. The authors plan to develop a more cognitively sound sequential model to capture a broader range of basic emotions and improve emotion classification accuracy. The paper concludes by outlining future research directions, including refining the feature set, improving parameter tuning, and exploring emotional intensity and transitions.This paper explores the *text-based emotion prediction* problem, aiming to classify the emotional content of sentences in children's fairy tales for text-to-speech synthesis. The authors use supervised machine learning with the SNoW learning architecture to classify emotional versus non-emotional content. Initial experiments on a dataset of 22 fairy tales show promising results, outperforming a naive baseline and a bag-of-words (BOW) approach. The study also discusses a tripartite model that considers emotional valence and explores different feature sets. The authors plan to develop a more cognitively sound sequential model to capture a broader range of basic emotions and improve emotion classification accuracy. The paper concludes by outlining future research directions, including refining the feature set, improving parameter tuning, and exploring emotional intensity and transitions.