The RapVerse Dataset: A New Benchmark in Text-to-Music and Motion Generation

This article surveys recent advancements in text-to-vocal, text-to-motion, and audio-to-motion generation models, highlighting the limitations of existing datasets and approaches. It introduces RapVerse, a novel dataset designed specifically for rap performance, pairing text, vocal, and motion data across 108 hours. Unlike prior work, RapVerse supports simultaneous generation of audio and full-body motion from text input, setting a new benchmark in multimodal AI for expressive performance synthesis