Given a target sound, finding the synthesizer preset that best reproduces it remains a core problem in sound design. Existing methods treat synthesis parameters as flat vectors, discarding the signal routing and parameter interactions that produce audio. We make two contributions. First, to learn a representation of parameters including their signal routing, we design a graph neural network whose message passing structure imitates FM signal processing. Second, we adapt the multimodal objective from SLAP to learn joint embeddings of audio and FM synthesizer parameters, enabling preset retrieval from a gallery. We focus on the Yamaha DX7, where six identical sinusoid operators interact according to one of 32 routing topologies. Our graph encoder's message passing weights are shared across all nodes and layers, enabling processing of arbitrary topologies of any size. When every topology is seen during training, the DX7-GNN and two baselines achieve strong audio-to-preset retrieval. When some topologies are held out for testing, the DX7-GNN substantially outperforms both baselines despite having the fewest parameters. Our ablations further support the claim that imitating FM signal flow in a parameter encoder improves generalization to unseen topologies.
Figures & tables
Figure 1: SLAP architecture for FM-SynAPSE: Online encoders ( EA , EP ) receive gradients; Target encoders ( EˉA , EˉP ) are updated via exponential moving average (EMA). Predictor networks ( PA,PP ) map online projections to predictions of target projections. Four cosine similarity losses combine intermodal ( LA→P , LP→A ) and intramodal ( LA , LP ) alignment. Dotted lines indicate how loss gradients backpropagate towards predictions, not target projections.
Figure 2: Two views of Algorithm 12: (Left) Traditional DSP. (Right) A message passing graph. Inactive edges are dotted lines.
Figure 3: Distribution of 31,443 presets across the 32 DX7 algorithms, colored by the held out split: train (odd algorithms); validation and test (disjoint even algorithms). The count per algorithm ranges from 116 (Algorithm 21) to 3,758 (Algorithm 5).
Component
DX7-GNN
Transformer
Highway
Input / Embedding
667K
6,568K
3K
Blocks / Layers
1,331K
50,401K
6,191K
Output / Projection
1,379K
264K
394K
Total
3.38M
57.2M
6.59M
Table 1: Parameter counts for the three parameter encoders. All models share the same audio encoder (82.8M) and SLAP projector/predictor heads (3.4–3.6M per arm, depending on encoder output dimension).
Audio → Param
Mod. Gap
Model
R@1
R@10
MRR
( L2 Dist.)
DX7-GNN
86.1%
99.5%
0.917
0.0293
Transformer
86.1%
99.5%
0.917
0.0209
Highway
81.2%
98.5%
0.880
0.0268
Table 2: Architectural comparison with 80/10/10 random split across all 32 algorithms (3,144-preset gallery). Bold marks the best value per column.
Audio → Param
Mod. Gap
Model
R@1
R@10
MRR
( L2 Dist.)
DX7-GNN
52.2%
88.5%
0.652
0.0446
Transformer
34.6%
70.6%
0.470
0.0505
Highway
24.8%
59.1%
0.365
0.0722
DX7-GNN augmentation ablation
No Swap
13.5%
47.0%
0.245
0.0617
Table 3: Retrieval on held out algorithms (4,096-preset gallery). Bold marks the best value per column.
Figure 5: Learned monotonic remappings overlaid with DX7 ground truth. (Left) Output level: the learned curve tracks the DX7’s exponential gain response. (Right) Feedback edge weight, scaled by the 0.5 factor from Section 3.4.2 : the learned curve suppresses mid-range values, moving toward the DX7’s power-of-two staircase.
Model
yA
zA
qA
DX7-GNN
68.4%
69.3%
69.3%
Transformer
71.0%
71.3%
71.1%
Highway
68.3%
68.7%
68.1%
MFCC
68.7%
—
—
LAION-CLAP
71.8%
70.8%
—
PANNs (AudioSet)
61.4%
—
—
Table 4: Triplet agreement (%) on timbremetrics [ 28 ] . yA : encoder output; zA : projection; qA : prediction. Bold marks the best value per column.
Figure 6: t-SNE visualization of held out algorithm embeddings. (Left) Audio and parameter embeddings by color. (Right) Embeddings from both modalities colored by algorithm.
Sound matching can be formulated as optimizing synthesizer parameters against an audio-domain objective. However, objectives derived from generic audio representations are often difficult to optimize, while direct search requires rendering every candidate. We introduce Synth-JEPA, which learns mutually predictive audio and parameter representations from paired synthesizer data. At inference, candidate parameters are scored directly in this learned space, yielding a renderer-free objective whose audio geometry is shaped by parameter correspondences rather than generic audio similarity. We evaluate Synth-JEPA on Surge XT using held-out synthesizer sounds and out-of-domain NSynth and FSD50K targets, against inverse models, direct search, and learned proxy objectives. Synth-JEPA outperforms all baselines in-domain and remains competitive out-of-domain. Its matching quality continues to improve with additional test-time search, allowing compute to be traded for match quality. In pairwise listening tests, listeners preferred Synth-JEPA in 85% of trials overall. Together, these results show that an audio representation with a parameter-induced geometry allows synthesizer sound matching to be approached as an effective renderer-free search problem.
Ben Hayes, Haokun Tian, Stefan Lattner
Sony Computer Science Laboratories, Paris, France · Queen Mary University of London, UK
This paper introduces PEACE, the first joint embedding of audio effect code and output audio. Building on SLAP's multimodal objective, we pair an AFx-Rep audio encoder with two code encoders for Faust, a functional language for audio signal processing. First, we evaluate a fine-tuned T5 transformer over Faust source code. Second, we evaluate a message-passing graph neural network over an intermediate representation of the Faust compiler, capturing both topology and UI parameters. We evaluate on audio-to-code retrieval, where masking UI parameters at inference yields embeddings that encode effect chain topology alone. When parameters are visible, the two code encoders tie on retrieval of mixed-length chains but have tradeoffs on single-effect galleries. With parameters fully masked, PEACE recovers ordered chain topology far above chance without the limitations of supervised methods. PEACE outperforms pretrained models on an out-of-distribution reverb retrieval benchmark and can improve frozen audio-only representations. Its dual understanding of topology and parameters lays the groundwork for music information retrieval systems that search, generate, and condition on DSP code.
Existing audio-to-MIDI tools extract notes but discard the timbral characteristics that define an instrument's identity. We present Instrumental, a system that recovers continuous synthesizer parameters from audio by coupling a differentiable 28-parameter subtractive synthesizer with CMA-ES, a derivative-free evolutionary optimizer. We optimize a composite perceptual loss combining mel-scaled STFT, spectral centroid, and MFCC divergence, achieving a matching loss of 2.09 on real recorded audio. We systematically evaluate eight hypotheses for improving convergence and find that only parametric EQ boosting yields meaningful improvement. Our results show that CMA-ES outperforms gradient descent on this non-convex landscape, that more parameters do not monotonically improve matching, and that spectral analysis initialization accelerates convergence over random starts.