Preprint
Aug 2026
DDSynth-RL: Audio Synthesizer Inversion via Discrete Diffusion with Reinforcement Learning
Experiments on Dexed show that, after supervised training, the discrete diffusion model is competitive with autoregressive and flow-matching baselines, and reward-based fine-tuning further improves out-of-domain audio matching performance.
Tristan Wu, Daniel Chin, Ju-Nan Zhang et al.
· 1 citation
· ⚡1