A masked discrete diffusion model fine-tuned with GRPO on rendered-audio rewards improves out-of-domain synthesizer parameter estimation on the Dexed FM synthesizer.
2) Parameter-space losses often fail to reflect rendered audio similarity, while the synthesizer be- ing a non-differentiable black box prevents simple audio- domain supervision
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.SD 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
DDSynth-RL: Audio Synthesizer Inversion via Discrete Diffusion with Reinforcement Learning
A masked discrete diffusion model fine-tuned with GRPO on rendered-audio rewards improves out-of-domain synthesizer parameter estimation on the Dexed FM synthesizer.