Sketch Controls for Synth Presets

Abstract

Sound matching and preset editing both ask how synthesizer controls can reproduce the audible shape of a reference. The current browser demo uses a conditional flow-matching model: reference audio plus either extracted or authored music-sketch controls generate 92 encoded coordinates, decoded to 89 Surge XT synthesizer parameters. Surge then renders the prediction locally in the browser.

This is parameter generation, not direct conditioning on a source preset and not a minimal-change editor. A planned preset-editing workflow will accept an FXP, render it in the browser, and use that audio as the inference reference. FXP upload and FXP export are not available in the current demo; current downloads are rendered WAV audio and a JSON evaluation record.

System overview forthcoming Reference audio + extracted or authored controls → flow inference → Surge parameters → browser render

Figure 1. Current demo path. The future FXP workflow will enter through browser rendering, not by directly encoding the uploaded preset's parameter vector.

Current interactive system

Live Demo

Open standalone evaluator

The evaluator runs ONNX inference and Surge XT rendering in this page. Audio stays in your browser. Loading is opt-in because the model and engine require about 200 MB of downloads. A short, low-step run is an integration smoke test, not evidence of sound-matching quality.

Model and interface

Method

Reference representation

The model consumes stereo mel features from the reference audio and 386 music controls on 32 frames. Controls can be extracted from separate sketch audio or authored as a pitch activation contour with loudness, spectral-centroid, onset, and end values.

Flow inference

Classifier-free guidance can use content and sketch together or independently. A seeded RK4 sampler predicts the encoded parameter coordinates, which are decoded into a Surge patch and rendered by the browser engine.

Current evaluation

The page reports integration diagnostics including normalized mel MAE, waveform RMSE, RMS levels, and peak amplitude. These measurements check a run's wiring and output; they are not a published perceptual evaluation or a preset-preservation result.

Pitch-control figure forthcoming

Loudness-control figure forthcoming

Centroid-control figure forthcoming

Flow trajectory figure forthcoming

Figure 2. Planned control-track and trajectory visualizations. No placeholder is presented as an experimental result.

Applications and scope

Sound matching now, preset editing planned

Available: sound matching

Upload reference audio, extract or author controls, generate a patch, listen to the real Surge browser render, and download WAV and JSON artifacts.

Planned: preset editing

Uploading FXP will render that preset first and use the rendered audio as reference. Direct source-parameter conditioning, minimal-change guarantees, and FXP download are outside the current system.

Scientific material

Listening samples and poster PDF

Curated listening samples, control-track plots, and the final ADC 2026 poster PDF are forthcoming. This page deliberately provides no broken media controls or unverified results.

Limitations

The browser engine is an experimental Surge 1.4 port; training used SurgePy 1.3.4.
The demo is specific to Surge XT and four-second, 44.1 kHz stereo inputs.
Low-step smoke runs verify integration, not model quality or perceptual similarity.
The current model does not promise that a generated patch stays near an existing preset.

Presented at

Virtual poster, Audio Developer Conference (ADC) 2026.

@misc{tinubu2026sketchcontrols,
  title={Sketch Controls for Synth Presets},
  author={Khaled Tinubu},
  howpublished={Virtual poster, Audio Developer Conference (ADC)},
  year={2026},
  url={https://synth.tinaudio.com/}
}