TableTalk Narrative Voice Classification and Retrieval for Tabletop RPGs

This project builds a modular audio pipeline for tabletop RPG session analysis within the HumanAI TableTalk framework. It extends mood detection with speaker diarization, narrative role classification, and semantic retrieval over recorded sessions. Deliverables include a Whisper-based transcription layer, an SVM/ensemble classifier for narrative voice tagging, a FAISS-backed semantic retrieval index, and a documented evaluation suite; enabling players to query sessions by emotional arc or character voice.