Search papers, labs, and topics across Lattice.
This paper tackles the complex problem of synthesizer inversion by framing it as a conditional generation task over discrete synthesizer parameters using masked discrete diffusion. The authors address the challenges of non-differentiable synthesizers and the inadequacy of parameter-space losses by incorporating audio-domain rewards through GRPO-style fine-tuning. Experimental results demonstrate that their approach not only matches but also surpasses the performance of existing autoregressive and flow-matching models in out-of-domain audio matching.
Reward-based fine-tuning of a discrete diffusion model significantly enhances synthesizer inversion performance, outperforming traditional methods in audio matching tasks.
Synthesizer inversion is challenging for two main reasons: 1) Distinct parameter configurations can produce perceptually similar sounds. 2) Parameter-space losses often fail to reflect rendered audio similarity, while the synthesizer being a non-differentiable black box prevents simple audio-domain supervision. To address the one-to-many mapping induced by the first challenge, we formulate synthesizer inversion as conditional generation over discrete synthesizer parameters and use masked discrete diffusion as the generator. This treatment additionally avoids the fixed-order assumption of autoregressive models and the continuous-relaxation mismatch of flow matching when modeling categorical synthesizer controls. To address the second challenge, we further fine-tune the model with GRPO-style audio-domain rewards computed from rendered outputs. 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. Code and demos are available at: https://github.com/DDSynth-RL/DDSynthRL.
