reward-conditioning

This paper presents a text-to-music generation system that leverages reward conditioning, expert iteration, and preference tuning to improve audio quality within a 120M-parameter model, submitted to the ATTM Grand Challenge at ICME 2026.