maolandaw/FCPE-burn

FCPE (Burn Format)

This repository contains Burn-format weights for the upstream model:

• CNChTu/FCPE (Fast Context-based Pitch Estimation, torchfcpe)

The published artifact is packaged as a Burn Pack (.bpk) archive. The repository also includes Rust tooling to regenerate the raw export manifest and Burn pack from the upstream checkpoint.

Repository Contents

Published model asset at the repository root:

• fcpe.bpk

Conversion and packaging sources included in the repository:

• convert.sh: one-step export-and-pack script that writes fcpe.bpk at the repository root
• src/bin/export_fcpe_raw.rs: exports the upstream ONNX checkpoint (fcpe.onnx) into .npy tensors plus a manifest file
• src/bin/check_fcpe_bpk.rs: loads a Burn Pack archive and prints the tensor count and total tensor bytes
• src/main.rs: packs exported tensors into Burn Pack archives
• Cargo.toml, Cargo.lock: Rust dependencies for the packer

Included Model

FCPE (Fast Context-based Pitch Estimation)

• Neural F0/pitch detection for singing voice
• Conformer/ConvNeXt-style encoder operating on mel spectrograms
• Outputs per-frame pitch probability distributions, decoded with local_argmax

The current FCPE export manifest contains 76 tensors (65 float32, 11 int64), all exported in FP32.

File Sizes

Current repository payload:

• fcpe.bpk: 43,317,528 bytes

Rebuilding the Artifact

The repository includes the script needed to regenerate the Burn artifact from the original upstream checkpoint.

Prerequisites

• A Rust toolchain capable of building the packer in this repository
• A local copy of the upstream checkpoint arranged like this:

CKPT_ROOT/
└── FCPE/
    └── fcpe.onnx

The upstream torchfcpe package bundles the same weights as torchfcpe/assets/fcpe_c_v001.pt (a PyTorch pickle, not parseable with pure Rust). An ONNX export of the identical graph (fcpe.onnx) is vendored in the niobures/FCPE Hugging Face mirror of the FCPE repository and is the export source used here. The exporter reads the ONNX initializers directly — no Python, PyTorch, or ONNX Runtime is required.

One-Step Conversion

./convert.sh CKPT_ROOT

This regenerates fcpe.bpk at the repository root and removes the temporary artifact directory when finished.

Download the Upstream Files

mkdir -p CKPT_ROOT/FCPE
curl -L -o CKPT_ROOT/FCPE/fcpe.onnx \
  https://huggingface.co/niobures/FCPE/resolve/main/onnx/fcpe.onnx

Export Raw Tensors and Manifest

If you want to run the steps manually instead of using ./convert.sh:

Example:

cargo run --release --bin export_fcpe_raw -- \
  --checkpoint-root CKPT_ROOT \
  --fcpe-subdir FCPE \
  --output-root artifacts/fcpe

This produces:

• artifacts/fcpe/fcpe_raw_f32/

The raw directory contains .npy tensor files and a manifest.json.

Pack Burn Archive

The Rust packer reads a manifest and writes a .bpk archive:

cargo run --release --bin fcpe-burn -- \
  --manifest artifacts/fcpe/fcpe_raw_f32/manifest.json \
  --output artifacts/fcpe/fcpe.bpk

Verify the Archive

cargo run --release --bin check_fcpe_bpk -- fcpe.bpk

Notes

• This repository is a model artifact and conversion repo, not a complete inference application.
• The generated FP32 manifest references FP32 tensors; the 11 integer tensors are shapes/constants kept as int64.
• The repository includes the large Burn archive directly, so Git LFS configuration in .gitattributes is part of the expected publishable layout.

License

This conversion is distributed under the MIT License, matching the upstream CNChTu/FCPE license. Upstream copyright: (c) 2023 CN_ChiTu. See LICENSE for the full text.