This project turns spoken input into an urgency score using speech recognition and a language model. We learned how transformer models process text, how regression differs from classification, and how to connect multiple AI components into one pipeline. We built the system by converting audio to text with Whisper and then passing that text into a RoBERTa model that predicts a score from 1 to 10. One challenge was adapting a language model to output a continuous value instead of categories, along with managing model performance and data quality.
We were inspired to create this project to help reduce long wait times for crisis hotlines and better manage incoming requests. Nonprofit organizations are often understaffed and experience high traffic, so a system like this could help prioritize cases more efficiently.
Log in or sign up for Devpost to join the conversation.