Skip to content
Other

PawCT and PagCT

choral singing to MIDI transcription

2 min readupdated

A pair of PyTorch checkpoints by Hanyu Meng for transcribing mixed-choir recordings into MIDI sheet music. PagCT merges all voices into a single combined track, while PawCT base splits the score into four separate parts: soprano, alto, tenor, and bass (SATB). The models tackle AMT (Automatic Music Transcription) with singing voice assignment. This tool is built for musicians, arrangers, choir directors, and MIR (Music Information Retrieval) researchers.

Key features

  • Two output modes: a combined track in pagct.pth or separate SATB tracks in pawct-base.pth.
  • Ready-to-use JSON configs (pagct.json and pawct-base.json) with tuned thresholds for frame, onset, and offset detection.
  • End-to-end training with union supervision and a learned presence gate for PawCT.
  • Verified reproducibility on the official YouChorale test split (30 recordings).

What you can do

  • Digitize an archival or rehearsal recording of a mixed choir into an editable MIDI file for further work in a notation editor or DAW.
  • Extract individual soprano, alto, tenor, and bass parts using PawCT base to quickly prepare rehearsal backing tracks or sheet music for choir singers.

Example

settings = json.loads(Path(f"{model_id}.json").read_text(encoding="utf-8"))
cfg = OmegaConf.merge(OmegaConf.load(checkout / "src" / "config.yaml"), settings["public_code_overrides"])
audio, _ = librosa.load(audio_path, sr=cfg.feature.sample_rate, mono=True)
transcriber = ChoralAMTTranscriber(cfg, settings["checkpoint_file"])
probabilities = transcriber.transcribe(audio)["output_dict"]
write_satb_midi(tracks, Path(f"{model_id}.mid"))
Read moreDetails
YouChorale note F1 @ 50 ms metric (accounting for pitch and onset within 50 ms): 0.3790 for PagCT and 0.2248 for PawCT base.
Running the models requires the Paw Choral AMT repository (revision b0718f8), librosa, and omegaconf.
Based on the paper "Toward Part-Aware Choral Transcription with Singing Voice Assignment" (arXiv:2610.09861).
Discuss Telegram 1

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 0.1 GB · ready files0.6 GB
  • FP8 / INT8recommendedweights 0.0 GB0.5 GB
  • 4-bitweights 0.0 GB0.5 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 2060 · BF16 — from RTX 2060

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # Hanyu-Meng/PawCT-PagCT
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download Hanyu-Meng/PawCT-PagCT --local-dir ~/models/PawCT-PagCT
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to PawCT-PagCT
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

New posts about versions of PawCT-PagCT (quants, finetunes, LoRAs) will show up under the bell.

No known relatives of this model yet. All derivatives on Hugging Face

About the model

Hanyu-Meng/PawCT-PagCT
Author
Hanyu-Meng
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

01

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments

Other

GyroFlow

GyroFlow