Gradient-based optimisation of modulation effects

Alistair Carson, Alec Wright and Stefan Bilbao

Acoustics and Audio Group
University of Edinburgh
Edinburgh, UK

🔊 Audio Examples         🗞️ Paper         🎛️ Dataset

Inference

Run inference using a provided pre-trained model:

python3 inference.py --model_path weights/BF-2-A/qa6tmyhk --in_audio your_file.wav

where --model_path can be replaced with another from the weights/ directory, and your_file.wav can be replaced to a file of your choice.

Dataset

The dataset is available here: https://zenodo.org/records/18186611. Download and extract the files into modulation_dataset directory, or update DATASET_PATH in config.py accordingly.

Train models

Train a model on a given configuration:

python3 main.py --config 0

where the config index can be chosen from the table below (see config.py).

Effect configurations

The tables below show the parameter configurations for the 3 effects pedals considered.

BF-2 Flanger
Config	Manual	Depth	Rate	Feedback	Label
0	0	1	0.5	0	BF-2-A
1	0	1	0.5	0.5	-
2	0	1	0.5	1	BF-2-B
3	0	0.5	0.5	0	-
4	0	0.5	0.5	0.5	-

SV-1 Chorus
Config	Speed	Depth	Wave	Filter	Label
10	0.5	0.5	0	1	SV-1-A
11	0.5	0.5	0.5	1	-
12	0.5	0.5	1	1	SV-1-B
13	0.5	1	0.5	1	-
14	0.5	1	1	1	SV-1-C

Smallstone Phaser
Config	Rate	Color	Label
20	0.75	0	-
21	0.5	0	SS-A
22	0.25	0	-
23	0.75	1	SS-B
24	0.5	1	-

Citation

Is using this work please use the following citation:

    @article{Carson2026modfx,
    title={Gradient-based optimization of modulation effects},
    author={Carson, Alistair and Wright, Alec and Bilbao, Stefan},
    year={2026},
    month=jul,
    journal={Journal of the Audio Engineering Society},
    volume={74},
    pages={533--543},
    }