vst-bench-ml

Generate audio-plugin datasets with records of how each output was produced. vst-bench-ml samples plugin parameters, coordinates offline rendering, records software and render settings, and checks stored artifacts for changes or missing files.

Install

Use Python 3.12 in a virtual environment:

python3.12 -m venv .venv
source .venv/bin/activate
python -m pip install vst-bench-ml==0.1.1
vst-bench-ml --help

Alternatively, install a source checkout with python -m pip install ..

The metadata permits Python 3.10 or newer. The documented rendering environment is macOS arm64 with Python 3.12; see the installation guide for the tested scope.

First dataset

Rendering needs a separately installed host and a plugin. The sample-free companion walkthrough provides both and takes you from installation to eight small generated outputs. The broader vst-bench MCP testing application is not required.

For a configured compatible host:

vst-bench-ml render --plugin /absolute/path/to/example.vst3 \
  --param-manifest parameter-manifest.json --n 4 --out output \
  --seed-label example-v1
vst-bench-ml verify --manifest output/manifest.json

Use a new output directory and a fresh host for each render. Offline verification needs only this Python package and the stored dataset, without running a plugin.

Learn more

• User documentation
• Installation
• Commands and outputs
• Troubleshooting
• Architecture and host protocol
• Issues and contributing

The package contains Python tools; the host, plugin binaries, datasets, presets, and audio are supplied or generated separately. Hashes establish artifact integrity. Matching output bytes have only been demonstrated on pinned environments; scientific suitability and cross-platform audio identity require separate evaluation.

Source, licence, and citation

MIT source; dependency notices are in NOTICE. The 0.1.0 Zenodo archive remains the historical release. See public source history and CITATION.cff for version-specific attribution.