SwiftF0 (safetensors)

Compact CNN pitch tracker from Nieradzik, 2025 (arXiv:2508.18440). Original implementation: lars76/swift-f0.

This repository hosts the released SwiftF0 weights converted from the upstream ONNX model to safetensors for use with the candle ML framework via candle-swift-f0.

Files

File Approx. size
swift-f0.safetensors 375 KB

The bundled ONNX is already inference-fused (Conv plus ReLU, no BatchNorm), so the conversion is a pure rename pass that preserves the original tensor shapes.

Names:

conv{i}.{weight,bias}           i in 1..=5
freq_projection.{weight,bias}

Provenance

Converted from the bundled model.onnx of the swift-f0 PyPI package using scripts/export_safetensors.py.

Parity

The candle implementation reproduces the reference ONNX forward pass to within 1e-4 max absolute difference on logits and on decoded pitch. Verification runs in scripts/pytorch_parity.py and in the Rust integration tests under candle-swift-f0/tests/.

Citation

@misc{nieradzik2025swiftf0,
  title={SwiftF0: Fast and Accurate Monophonic Pitch Detection},
  author={Lars Nieradzik},
  year={2025},
  eprint={2508.18440},
  archivePrefix={arXiv},
  primaryClass={cs.SD},
  url={https://arxiv.org/abs/2508.18440}
}

License

Same as upstream SwiftF0: MIT, Copyright (c) 2025 Lars Nieradzik. See LICENSE.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Paper for jhartquist/swift-f0