Smart Story Telling Model with Emotion-Based Enunciation and an Interactive Query Resolver

Emotion detection from text is a field of research that involves the recognition of emotions such as anger, disgust, fear, happiness, sadness, and surprise from text. The conversion of text to audio with enunciation based on the detected emotions from text is a contemporary field of study that has short explorations done. The paper presents a methodology and design to classify the emotional affinity of sentences in the narrative domain of children’s fairytales using a transformer-based natural language processing technique called the BERT model for subsequent usage in the appropriate expressive rendering of text to speech synthesis using statistical models. The paper also presents a methodology to implement a question-answering system to resolve queries regarding the story by producing an audio clip containing the most appropriate answer extracted through a natural language processing technique called the ALBERT model. The model developed can detect emotions from text with an accuracy of 83%, convert text to audio with emotion-based enunciation with an accuracy of 90%, and the question-answering system produces an accuracy of 80%. The future work will focus on increasing the features taken into consideration for manipulating neutral audio. Detection of sarcasm and handling of complex sentences having multiple emotions will also be included.